You are on page 1of 182


This page intentionally left blank
Knowledge and
the State of Nature
An Essay in Conceptual Synthesis


This book has been printed digitally and produced in a standard specification
in order to ensure its continuing availability


Great Clarendon Street, Oxford 0X2 6DP

Oxford University Press is a department of the University of Oxford.
It furthers the University's objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Bangkok Buenos Aires Cape Town Chennai
Dar es Salaam Delhi Hong Kong Istanbul Karachi Kolkata
Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi
Sao Paulo Shanghai Singapore Taipei Tokyo Toronto
with an associated company in Berlin
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
Edward Craig 1990
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
Reprinted 2002
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose this same condition on any acquirer
ISBN 0-19-824243-3
ISBN 0-19-823879-7 (Pbk)
This page intentionally left blank

I am grateful to many students, colleagues, and friends from various

universities in Britain and Germany for their interest in and com-
ments on earlier versions of parts of this monograph. My principal
debts can be gathered from the text itself. A number of people have
given me the benefit of their native-speaker's expertise over the
linguistic data that I refer to in Sections XVI and XVII. Their gener-
ous help is much appreciated.
Particular thanks are due to the Leverhulme Trust for enabling
me to devote a large part of the year 1986-7 to this project. How
much longer it would have taken to reach the present state without
such assistance, or whether it would ever have been reached at all,
I can only guess.
Finally, my thanks to the editor of the Proceedings of the Aristote-
lian Society for permission to re-use material which appeared in
Churchill College, Cambridge E.J.C.
This page intentionally left blank

I Nature and motivation of project. Doubts

answered. Plato, Pears, Hobbes, comparison with
State-of-Nature Theory in Political Philosophy.
Evolutionary epistemology. 1
II Derivation of first condition; the problem whether
belief necessary. Necessary and sufficient
conditions an unsuitable format. The prototypical
case. 11
III Need for third condition. Discussion of the
NozickDretske analysis. 18
IV Why causal theory, tracking, reliabilism all good
approximations. Why justified true belief a good
approximation. Comparison with Grice. 24
V Distinction between Informant and Source of
Information; its nature and point. Application to
putative knowledge without belief cases; and to
comparativism: Goldman. 35
VI Being right by accident. All analyses insufficient.
Blackburn: the Mirv/Pirv principle. 45
VII Local v. Global Reliabilism. Discussion of
McGinn. 54
VIII Externalist and Internalist analyses. The first-
person case. Knowing that one knows. 61
IX Insufficiency of the various analyses. The No false
lemma principle. Its rationaleand its effect. 69
X Objectivisation. The cart before the horse
objectionand the response. 82
XI Lotteries and multiple premises: the pull towards
certainty. Knowledge and natural laws. 98
XII Objectivisation and scepticism. Ungers first
account. 104
XIII Two explanations of scepticism: the first-person
approach, and the absolute perspective. 120

XIV Knowledge and involvement. What makes truth

valuable? 130
XV Testimony and the transmission of knowledge.
Welbourne: believing the speaker. 134
XVI Other locutions: Knowing Fred. Information v.
acquaintance. Interacting with Fred. Knowing
Londonand German. 140
XVII Other locutions: Knowing how to. The Inquirer
and the Apprentice. Knows how to compared
with canand with knows that. 149
Appendix Ungers Semantic Relativism 162
References 168
Index of Names 171

The standard approach to questions about the concept of knowledge

has for some time consisted in attempts to analyse the everyday
meaning of the word know and its cognates. Such attempts have
usually taken the form of a search for necessary and sufficient con-
ditions which, when measured against our reactions to examples
both real and imaginary, match our intuitive ascriptions and with-
holdings of the title of knowledge. We are to provide, if you like,
an explicit intension to fit the intuitive extension.
One might wonder whether, if the idea is to analyse the concept
of knowledge, this can really be the right programme. As well as
intuitions about the extension of the concept, we seem also to have
certain intuitions about its intension, that is to say intuitions about
why certain cases do, and others do not, qualify as knowledge.
Thus we may feel about a certain example, both that the subject
does not have knowledge, and that he does not have it because
the truth of his belief is accidental (for instance). The sceptic notor-
iously tries to show that the two do not mesh: our intuitions about
the intension, the conditions of application of the concept, in fact
determine a much smaller extension than that which our directly
extensional intuitions mark out. If he is wrong, the point needs
arguing; if he is right, the question arises: to which set of intuitions
should we give priority in order to arrive at the analysis of the
everyday concept ? Either way, a good deal of work in epistemology
and the theory of meaning (which in the light of history one can
hardly expect to be uncontroversial) must be done or assumed just
to reach the stage of saying that there is such a thing as the everyday
concept of knowledge at all, let alone settle any question as to how
one should proceed to analyse it. So if the standard approach runs
into difficultiesand the work of the last twenty-five years makes
it apparent that it doesit is surely worthwhile to try to think
of another.
And there is another problem, though in this case it may be less

a flaw in the approach itself than a defect in the attitude commonly

taken towards it. Let us suppose, however optimistically, that the
problem of the analysis of the everyday meaning of know had both
been shown to exist and subsequently solved, so that agreed neces-
sary and sufficient conditions for the ascription of knowledge were
now on the table. That would be a considerable technical achieve-
ment, and no doubt a long round of hearty applause would be in
order, but I hope that philosophers would not regard it as a terminus,
as many writers make one feel they would. I should like it to be
seen as a prolegomenon to a further inquiry: why has a concept
demarcated by those conditions enjoyed such widespread use? There
seems to be no known language in which sentences using know
do not find a comfortable and colloquial equivalent. The implication
is that it answers to some very general needs of human life and
thought, and it would surely be interesting to know which and
how. And the threat, of which some writers have seemed largely
unaware, is that the more complex the analyses preferred in response
to the flood of ingenious counterexamples (and some are very com-
plex indeed), the harder that question will be to answer.
These two thoughts, that it will do no harm to have an alternative
angle on the concept of knowledge that does not start from its sup-
posed extension, and that its purpose should be at least as interesting
as its analysis, together motivate an experiment. Instead of beginning
with ordinary usage, we begin with an ordinary situation. We take
some prima facie plausible hypothesis about what the concept of
knowledge does for us, what its role in our life might be, and then
ask what a concept having that role would be like, what conditions
would govern its application. Such an investigation would still have
an anchorage point in the everyday concept: should it reach a result
quite different from the intuitive intension, or one that yielded an
extension quite different from the intuitive extension, then, barring
some special and especially plausible explanation of the mismatch,
the original hypothesis about the role that the concept plays in our
life would of course be the first casualty. For it is not the idea to
construct an imaginary concept, but to illuminate the one we actually
have, though it be vague or even inconsistent; and to illuminate
it by showing that a concept with the hypothesised role would have
characteristics closely resembling those that it exhibits itself. But
should our intuitions prove indeterminate or elastic, this type of
investigation might reveal constructive ways of stretching them, and

the rationale behind the stretch. With luck it might also reveal the
sources of the indeterminacy or elasticity which dogged the attempts
to answer, or even to ask, the first familiar question.
It can at least be said for this way of creeping up on the concept
of knowledge that we are asking a question that can reasonably
be expected to have an answer. One doesnt have to commit oneself
to a great deal of epistemology or semantic theory, as the standard
approach evidently does, to presume that there is such a thing as
the point of this concept, what it does for us, the role it plays in
our lives. And if this is so, one way to find out must be to form
some hypothesis about it, try to work out how a concept custom-
designed for that role would look, and then see to what extent it
matches our everyday practice with the concept of knowledge as
actually found. We may then have to revise or supplement the
hypothesis from which we began, but that will hardly be surprising,
and certainly no cause for instant despair.
Whilst agreeing that we may expect the concept to serve some
purpose, however, we might doubt whether the consideration of
its purpose will necessarily lead to anything like an analysis, or
to anything that can be measured against the intuitive extension.
We might doubt, in fact, whether it will necessarily lead anywhere
interesting at all. Every language, an objector might reason, has
a word for water. And having that word has an important purpose,
namely, to make it possible to talk about water, something which
every community has an obvious need to be able to talk about.
But no a priori thought about that purpose will bring us any closer
to an analysis of the concept of water, even if the notion of an analysis
be very generously interpreted. Couldn't it just be that knowledge,
like water, is common and important stuff, and that the purpose
of the concept is simply to enable us to think and talk about it?
Though I would be hard put to it to argue the point, I am fairly
confident that this is mistaken. Knowledge is not a given phenome-
non, but something that we delineate by operating with a concept
which we create in answer to certain needs, or in pursuit of certain
ideals. The concept of water, on the other hand, is determined by
the nature of water itself and our experience of it. But probably
a better response here, at any rate a less dogmatic one, is the proof
of the pudding: if some hypothetical but plausible purpose does
issue in conditions of application showing a close fit to the intuitive
extension of know, and does fit well with a variety of facts about

the phenomenology of the concept, then those who hold it to be

mere coincidence may be requested to make a case for their attitude.
Another objection would be that the purpose, or purposes, of
the concept of knowledge, though no doubt there are such things,
are most unlikely to be anything so simple as the ones considered
in this essay. The method presented here ties itself to purposes of
a severely practical kind arising in what might be called a primitive
situation. Suppose (and isn't it really more likely?) that the concept
comes into existence in response to rather more sophisticated levels
of consciousness?
With that I would initially have been inclined to agree, though
I am not sure where the inclination comes fromthere may just
be an intellectual prejudice to the effect that everything must really
be frightfully complex. Certainly there is one thing we shouldn't
say in reply to this objection: that the concept under investigation
is so widespread, so ancient, that it must have its origin in the most
primitive requirements. What may well follow is that it must have
its origin in primitive societies, but there is no guarantee whatever
that primitive societies have only primitive requirements. Any
society that has a well-developed language, sufficiently well deve-
loped for us to be able to say that it exercises a concept even approxi-
mately identifiable with our concept of knowledge, consists of
creatures that have reached a considerable degree of mental com-
plexity. Any number of different sorts of need may, for all we know
to the contrary, follow in the wake of this complexity; so there
is no a priori reason to think that we are tied by methodological
principles to considering only needs of the very basic kind that I
have actually tried to restrict myself to.
Again, the best response will be to treat our strategy as a hypothe-
sis. If it doesn't work, doesn't issue in a concept having at least
very close similarity to the concept we are explicating, then we
shall have to modify the hypothesis and propose one involving rather
more advanced features of human consciousnessbut there is no
reason to bring them in before the progress, or lack of progress,
of the investigation makes it necessary to do so. To start off without
them doesn't mean making the assumption that they will not in
the end be needed; it is no more than good method to test the
explanatory powers of the simple before resorting to the complex.
Something similar will apply to the question of what further
needs it will be legitimate to introduce, should that prove necessary.

We shall be looking for features of human psychology which may

plausibly be supposed to be possessed by all humans, preferably
ones which there is some independent reason to suppose to be pos-
sessed by all humans. To illustrate the point with an example of
one which cannot without qualms be thought of in this way, we
might suggest the wish to explain, in some fashion, the behaviour
of one's fellows, or the wish to understand them in a way which
makes them the same sort of being as oneself. (It might be thought,
and has been suggested to me, that this idea could help us to see
the concept of knowledge as some sort of theoretical construct, useful
for explaining why other members of our community behave as
they do.) But just how widespread this concern with explanation
is, in particular whether it is widespread enough to fit our present
bill, is very hard to saythinking in these terms might just be a
reflection of our contemporary obsession with the methods of the
natural sciences. For that reason alonethere may be othersit
would not be advisable to allow ourselves such a starting point before
we are sure that we have exhausted the potential of far less conten-
tious claims about the human situation, like those which I have
actually tried to exploit.
We should beware of an assumption which is easy to make, but
which could prove restrictive or even worse: that when we render
the final account of the concept which we have constructed, this
account must go comfortably into the form of a list of conditions
that are individually necessary and jointly sufficient for its correct
application. The standard approach sets itself the problem in this
form, but whether the most illuminating response to the request
for specification of the concept of knowledge has to fit it ought
to be regarded as an open question, especially if we are to approach
it from this different angle. In fact, I shall shortly be arguing that
in trying to fit our results to it we are more likely to switch the
illumination off. On the other hand, one clear fact about this matter
is that the concept does lend itself at least fairly well to treatment
in the standard format; and that is one of the facts which an illuminat-
ing account of it ought to illuminate, whether or not it fits that
format itself. And whilst we are about it we might aspire not just
to illuminate that fact, but also to cast light on the appeal of the
various particular analyses that have been advocated in the literature
of the topic.
I have spoken so far as if there really were something which I

have called the intuitive extension of the concept of knowledge,

and that the problems lay in deciding whether or not it matches
the intuitive intension, and what to do about it if it doesn't. But
the difficulty may go one layer deeper: is it so clear that there is
such a thing as the intuitive extension at all?
To say that there is might make things sound like this: we can
describe cases in terms of the facts, the subject's beliefs, the subject's
relation to the facts, his mental state, and so on. Then there will
be, amongst speakers of English of a reasonable level of intelligence
and education who can be persuaded to give the matter their undi-
vided attention, unanimity as to which cases exemplify knowledge;
the decision, in an individual case, will depend only on the features
of the case as described, and will not vary from umpire to umpire,
nor with the same umpire from time to time. If that is what is implied
it has to be said that it is very doubtful. There is at least some
reason to think that our judgement varies with the circumstances
in which we are invited to consider the case. If we come to believe
that Fred has an identical twin who was in town that day, our
decision that we knew that it was Fred catching the bus on the
other side of the street may change. Our attitude to a lot of know-
ledge claims may change if we are asked to consider them with
the thought of the Cartesian demon looming over us. And if it doesn't
change, who is to say whether that is because we steadfastly hold
to the proper application of know in spite of the philosopher's
attempts to deflect us, or rather because we obstinately refuse to
pursue its proper application into these unaccustomed regions in
spite of the philosophers attempts to help us do so? Answers to
questions like these are usually accompanied by sparks from the
grinding of some philosophical axe; if we want to start from first
principles we would do much better to record the fact that these
sceptical deliberations produce different reactions, often combined
with a sense of puzzlement or irritation. Then we can take that
fact as a part of the data about the life of the concept of knowledge
and hope for an account which will allow us to assimilate it as it
is, rather than begin by distorting it in the name of some supposedly
well-defined extension which the concept is arbitrarily alleged to
have. Nor will we be tempted to pretend that we know where to
draw a line between natural and philosophical consideration of
It seems, therefore, that we may be in the happy position of being

able simultaneously to widen and strengthen the basis of the investi-

gation. To widen it, because we can include amongst our explananda
such facts as the various analyses of the concept of knowledge that
philosophers have given, the fact that controversy about certain of
their clauses seems difficult to resolve, the facts about the reactions
to sceptical proposals like that of Descartes, and all this be it noted
not instead of but as well as the extension (or whatever can uncon-
troversially be found of it) on which the standard approach uniquely
concentrates attention. And we can strengthen it, because these facts
have more of the nature of an undoctored given than those on
which the standard approach fixates - even if it be true that all data
are theory-laden, we have surely reduced the theoretical load.
The idea that it may be worth asking after the roots of the value
of knowledge when investigating the concept has an ancient prece-
dent. In Plato's Meno, Socrates wonders why we should be inter-
ested in anything beyond true belief: after all, if you believe that
this is the road to Larissa, and want to go there, then acting on
your belief will get you what you want just so long as it is true,
so why ask any more of it? His answer is that true beliefs are even
more valuable if they are stable, that is to say if we persist in holding
them; and this means, so he continues, that it is advantageous to
have good reasons, for then we will hold fast to the truth and not
be easily lured away to falsehood.1
Plato's contribution here is far from negligible. But I shall not
adopt it. Whether the stabilisation of true beliefs is important or
not depends on which beliefs we are considering, and the circum-
stances of the agentmany beliefs are required for the guidance
of single, one-off actions under circumstances which will not recur,
and once the particular occasion is past there is no obvious value
at all in their persistence. (I might now need a true belief about
the time; but that this belief should persist, so that tomorrow I
will still know what the time was today, at the moment when I
wanted to know it, may be of no interest to rne whatever.) Apart
from that, the possession of good reasons is not the only, nor neces-
sarily the best, way to stabilise beliefs: effect of early upbringing,
emotive ties or Humean psychological mechanisms may be just as
good, and better. And apart from that again, we shall see that stability

Plato, 97 ff. (Works are specified by the author's name and, where appropriate,
a number, as listed under References, pp. 168-9.)

is not the only, and perhaps not even the chief, thing that we may
want in a true belief.
Even if we confine ourselves to recent literature of the analytic
school the kind of approach I shall recommend is not without prece-
dent. David Pears has suggested something of the kind:
It often happens, as Aristotle saw, that understanding how a thing has
developed from primitive beginnings helps us to understand it in its deve-
loped form. Certainly this is true of knowledge. In this case the primitive
beginning is the ability to make a discriminating response to circumstances.
In the early stages the response will be a piece of overt behaviour. Later,
it may be internalised, and stored for future use. 2

This point, which Pears did not go on to press, may well be useful,
but its primary use is most likely to be that of shedding light on
the concept of belief, and maybe, if we add the idea of the overt
behaviour, when it arrives, being in some sense successful, that of
true belief. It will not in itself bring us very close to the concept
of knowledge as we have it, unless an adequate account of that con-
cept turns out (somewhat against the early odds, most of us will
feel) not to need, beyond true belief, any further apparatus. So we
can't rest content with Pears' remarks, though we can, and should,
take his hint.
I have elsewhere referred to this project as the practical explica-
tion of knowledge. The notion of an explication came into currency,
I believe, through Carnap. To explicate a concept was not exactly
to analyse it; it was to construct a new version of it satisfying certain
standards, with the proviso that to count as a new version of that
concept it had to emerge with many of its principal features intact.
The procedure suggested here is analogous; but it is concerned with
the practical rather than (as in Carnap) the theoretical aspect of the
concept. Hence the title of the earlier paper. But let it not mislead:
Carnaps intentions were normative, the establishment of the con-
cepts fit to form the rational basis of the unified science, whereas
mine are the more purely theoretical ones of shedding light on the
nature and origins of present practice. In that respect at least I shall
follow the traditional approach.
Whilst speaking of precedents, we should notice that my project
can claim membership of another tradition, one which spreads itself

D. F. Pears. ' E. J. Craig (1).

altogether wider than conceptual analysis, however widely that be

conceived. I refer to the tradition of naturalism, in which thinkers
see man, his behaviour and institutions, as natural facts to be under-
stood as the (broadly speaking causal) outcome of other natural
facts. What concepts we use, what linguistic practices are common
amongst us, these are special cases of input to the more general
naturalistic enterprise. Hume's treatment of the concept of causality
is one. Another, in a quite different area, is Hobbes account of
political and legal institutions: they are more or less adequate
attempts so to regulate the power-relationships between individuals
as to avoid the otherwise inevitable 'war of every man against every
man' and create stable circumstances in which humans, and human
society, may flourish. Robert Nozick has writtenand I think there
is meant to be a note of criticism in the remarkthat unlike political
philosophy epistemology has not started from a consideration of
the state of nature. 4 If that is true, and a defect, my project may
be thought of as one way to supply the lack.
Hobbes believed, of course, that normative consequences were
to be drawn: a particular kind of constitution ought to be adopted,
and in this respect other state-of-nature theorists such as Locke,
and more recently Nozick, have followed him. That is a different
matter; whether it results from my deliberations that we in any
sense ought to operate the concept of knowledge much as we do
is a question I shall not address, for unlike Hobbes's theme it is
not contentious. But it will, I believe, emerge that without it life,
whether or not brutish and nasty, would be a good deal more soli-
tary, and almost certainly shorter. Some may also wish to ask a
very different normative question: whether, if the concept of know-
ledge is to be developed or rendered more precise this ought to
be done in one way rather than another; clearly, there are parallels
in political theory. Again, I shall not offer an opinion; in any case,
unless we are told the purpose of such development we do well
to have no opinion to offer.
In spite of such differences, these points suggest that one might
view the whole of the following investigation in a different light:
as an adaptation to the theory of knowledge of a procedure tradition-
ally at home in ethics and politics. I began as if our theme were
the analysis of the concept of knowledge: we were to apply a different
R. Nozick, (1).
10 SECT. I

method to the same topic. Somewhat less parochially, we may think

of our starting point as state-of-nature theory: we are to apply similar
methods to a different object: not political, but conceptual and
linguistic institutions.
Finally, before we leave the job-description for the job itself, a
third connection is to be noted. 'Evolutionary Epistemology', now
widely so called, is a branch-arm of the prevailing stream of natura-
lism. It looks at our cognitive faculties as adaptive responses to
changing circumstances and changing needs for information. My
investigation is akin, though it differs in two respects. First, I am
not concerned, except peripherally, with any particular sensory or
inferential faculties, but with our concept of knowledge in general.
Secondly, I shall not treat its development diachronically, and that
is not just an omission: if what I shall say is along the right lines,
the core of the concept of knowledge is an outcome of certain very
general facts about the human situation; so general, indeed, that one
cannot imagine their changing whilst anything we can still recognise
as social life persists. Given those facts, and a modicum of self-
conscious awareness, the concept will appear; and for the same rea-
sons as caused it to appear, it will then stay. Our cognitive faculties
may change, and if they do then what we know will change; but
that, it will be seen, is not a reason to expect the concept of knowing
to change as well.

Such is the programme; we can now begin its execution. Fortunately

there is a firmly fixed point to start from. Human beings need true
beliefs about their environment, beliefs that can serve to guide their
actions to a successful outcome. That being so, they need sources
of information that will lead them to believe truths. They have 'on-
board' sources, eyes and ears, powers of reasoning, which give them
a primary stock of beliefs. It will be highly advantageous to them
if they can also tap the primary stocks of their fellowsthe tiger
that Fred can see and I can't may be after me and not Fredthat
is to say, if they act as informants for each other. On any issue,
some informants will be better than others, more likely to supply
a true belief. (Fred, who is up a tree, is more likely to tell me the
truth as to the whereabouts of the tiger than Mabel, who is in the
cave.) So any community may be presumed to have an interest in
evaluating sources of information; and in connection with that inter-
est certain concepts will be in use. The hypothesis I wish to try
out is that the concept of knowledge is one of them. To put it briefly
and roughly, the concept of knowledge is used to flag approved
sources of information.
I shall not for the moment be concerned with the evaluation of
what I have called 'on-board' sources. In the ordinary way we simply
take it that the beliefs they mediate are true. To find oneself in
possession of a belief on the question whether/; pre-empts inquiry;
to take a self-conscious look at one's own apparatus with the doubt
in mind that it may have delivered a falsehood calls for a considerable
degree of sophistication. Our investigation ought to start from the
position in which we as yet have no belief about p, want a true
belief about it one way or the other, and seek to get it from someone
else. (I do not mean to suggest that it is ever in principle impossible
for us to find out for ourselves; but in practice that will often be
a hopelessly inefficient way of going about it.) Our interest in our
own faculties as sound sources of information has a part to play,

since under certain circumstances that interest becomes acute, for

very good practical reasons; but it would not be good method to
begin with it.
Consider then the position of someone seeking information on
the point whether or not p. What does he want? In the first place,
he wants an informant who will tell him the truth on that question.
The informant, we may assume, will not in general tell him the
truth unless he (the informant) holds a true belief about it. (Cases
of people who, whilst not holding true beliefs, insincerely give 'infor-
mation' which is in fact true, are rare; and informants who do that
regularly are as good as non-existent.) So the inquirer wants an
informant such that:
Either p and he believes that p, or not-p and he believes that
This gives us a start on the analysis (or whatever we should at this
stage call it) of the concept which characterises the source of infor-
mation that the inquirer hopes to encounter. It is, obviously, very
closely related to the first two clauses of the traditional definition
of knowledge. I say 'closely related to' rather than 'identical with'
only because there has been a change of perspective. The traditional
analysis concentrates on the form 'X knows that p', whereas this
approach directs us in the first instance to 'X knows whether p'.
It is clear that the standard opening clause 'p is true' cannot feature
in the analysis of the latterwe need something like the dis|unctive
formula given above.
The relevant part of the traditional analysis, true belief, is that
which virtually all subsequent attempts to define knowledge have
taken over. But before we get into the contentious area of further
conditions we should pause to ask: how strong a conviction is in
question here, when we speak of the informant as believing that
pi Some analyses speak of certainty at this point, some of being
sure. But it has been argued that being certain, or sure, or confident,
is not a necessary condition for knowledge, and Colin Radford 1
has gone further and claimed that the belief condition can be dropped
altogether, since cases of knowledge of p can be found in which
the subject actually believes, if anything, that not-p. Does our
approach have anything to say on this score? It does, though what

' C. Radford.
S E C T . II 13

it offers is not a one-way-or-the-other decision; its contribution

is not to divide the area into black and white but rather to make
it comprehensible that this part of it should be grey.
In seeking information we are seeking to come by true beliefs;
we do not want just to have truths enunciated in our presence,
but we want to be brought to believe them, so that these beliefs,
since they are beliefs (and not mere entertainings) can guide our
actionsand guide them to success because they are true. We shall
therefore want as an informant someone who has the following
If he tells us that/?, we shall thereupon believe that/;.
Now because the confidence with which we believe things affects
the way we announce them, and the way we announce them affects
the likelihood that the audience will believe us, this may make it
look as if there could be some minimum level of confidence necessary
if the informant is to meet the inquirer's needs: for any lower level
he will sound so hesitant that the inquirer will probably not come
to believe what he says. To try to place a quantitative value on
that level would be silly, but it is probably not too silly to describe
it by using another vague term, and saying that if he is successfully
to induce the belief that p in his audience, he had better believe
it himself. That would make it appear that the concept we are con-
structing will indeed include belief as one of the conditions for a
good source of information.
But we are going too quicklythe matter is not so simple. Other
things which we believe about the informant will also play a part
in determining whether we believe what he tells us (or indeed, what
is very much the same thing, whether we select him as an informant
at all), and under certain circumstances they may well be overriding,
so that we have little tendency to believe even a very confidently
made claim, or are prepared to accept even a very diffident utterance.
Very briefly, what I have in mind is that if the informant satisfies
any condition which correlates wellas we believewith telling
the truth about p, he will be regarded as a good source. Confidence
is only a special case of this, and not even one that we are always
prepared to use. The point will become clearer later on, after the
consideration of the further conditionsfurther to true belief, that
iswhich we shall want the informant to meet. For the moment
we can best leave it at this: that since confidence is not essential

to performance as a source, the constructed concept will concede

some ground to those accounts of knowledge which play down the
belief requirement. On the other hand, it will give no support to
those which want to play it down to vanishing point. In the vast
majority of cases a good informant will believe the information he
gives, and will give it, what is more, precisely because he believes
itthe counter-examples, though perfectly genuine, are freakish.
One might add that on many matters, those who hold a belief at
all nearly always hold a true one, so that in these types of case
possession of a belief is itself a property which correlates well with
being right. These are facts about our world which cannot be without
influence on our concepts. So if our hypothesis is on the right track,
it is neither surprising that so many take belief to be essential to
knowledge, nor that some deny it, nor that many people's intuitions
leave them in the lurch at this point.
Here we see starkly a major disadvantage of the approach which
takes its sole target to be the listing of logically necessary and suffi-
cient conditions. If it can be argued that belief is not a necessary
condition for knowledge, then belief will make no appearance on
the final balance sheet. There is no place on it where the conceptual
accountant can present belief as a major component of the constel-
lation of thoughts which go into the practice of operating with 'know'
and its cognates. He can try to talk about it, just so long as the
audience is prepared to listen to such penphera, as of something
which very often accompanies knowledge. But when he is asked
for the real outcome of the business, the analysis, anything not
strictly a necessary condition simply vanishes without trace. Of all
its deep centrality nothing whatever remainsit could be as inciden-
tal as the fact that nearly all knowers are less than 150 years old.
That greater flexibility in the description of concepts is required,
is hardly a new point. Wittgenstein famously wrote of family resemb-
lance concepts, for which fixation on the format of necessary and
sufficient conditions leads either to triviality or error. But -we can
also see, I think, why this greater flexibility is required, or at the
least, why it is required in this case. We are asking not so much:
when is the ascription of a certain concept correct, but rather, why
is it applied? In freakish circumstances, a purpose may be achievable
in unusual ways - factors which would usually frustrate it may,
if other features of the situation are exceptional, do no damage,
factors which are usually vital may, abnormally, be dispensible.

There is, however, an expository problem: how is this flexibility

to be expressed? If we speak as if belief were not a component of
knowledge we do it an injustice, if we say that it is then we risk
being understood to say that it is a necessary condition. If, whenever
the relation between believing and knowing crops up, we try to
present the situation as it actually is, long-winded clumsiness will
engulf us. I can think of nothing better than to ask the reader for
a change of gear: what may look like an attempt to state necessary
conditions should rather be taken as part of the description of a
prototypical case, a case from which speakers and their audiences
will tolerate, in the right circumstances, varying degrees of deviation.
How much deviation, and under what circumstances, ought to be
related to the purpose behind the formation of the concept in ques-
tion. The prototypical description enshrines the features that effect
realisation of the purpose when things are going on as they nearly
always do.
It is important to notice that 'the way things nearly always go
on' enjoys special status in this context, not just numerical advantage.
We must never forget that the inquirer's situation is a practical one:
he must pick out the good informant, or decide whether to make
use of a volunteered statement. Now freakish cases, in which for
instance he would finish up with a true belief, though the informant
offer him a falsehood, or offer him a truth without believing it,
are not merely rare. It is also hard for the inquirer to detect that
he is in the presence of such a casewhich he must do, and do
confidently, if he is to make use of it; if it is possible at all it will
only be because he happens to be in possession of a good deal of
collateral information. To try to make a practice of detecting freakish
cases would mean incurring high costs in time and energy; and suc-
cessful detection would scarcely ever offer any benefit which could
not be had by finding a standard informant, or investigating for
oneself. In practice, therefore, it must be the standard or prototypical
case at which the inquirer's strategy is directed, so that one might
almost say that for practical purposes what the concept amounts
to is the essential description of the prototypical case.
The words 'almost' and 'for practical purposes' are not in that
sentence for nothing, however. One thing a proper account of a
conceptual practice ought to be able to explain is why prima facie
counter-examples to a proposed definition have (at least prima facie)
the feel of being such. What is it, for instance, that gives Radford's

examples some purchase? Not every account of the concept of know-

ledge will automatically be able to answer that question. The present
suggestion does so by reference to the hypothesised purpose under-
lying the practice. This purpose, in conjunction with a few platitudes
about the way in which human inquirers operate, generates a set
of descriptive conditions; what they describe is the prototypical case
in which the purpose is standardly achieved. A speaker is not taught
these conditions explicitly, any more than he is explicitly taught
the purpose, but when a philosopher presents them to him as logi-
cally necessary and sufficient for knowledge he is not hard to con-
vinceafter all, they apply to all the cases that readily come to
his mind, and it is perfectly understandable that they should. The
exotic cases, on the other hand, in which he recognises the fulfilment
of the familiar purpose in the absence of one of these conditions,
pull him for that reason to acknowledge them as positive instances,
though very unusual ones.
Is he to go along with the pull, or is he to resist it? The practice
into which he has grown up cannot be expected to help him on
that point. Why should it? It has no need to legislate for cases with
highly unusual features, or rather highly unusual combinations of
features. No need, and in a sense it has no way to do it either.
It is precisely by being everyday practice that everyday practice
manages to impress itself upon speakers and so stay what it is. How
can it help in freakish, perhaps wholly imaginary, circumstances
in which some of the familiar indicators fall one way, some another?
And if it could, why bother? We can always resort to expressions
like 'It's almost as if he knew'.
I cannot, of course, generate the slightest appearance of a counter-
example to the sufficiency of an analysis of knowledge by postulating
a human subject who will be 207 years old next birthday, whom
I however allow to satisfy all the conditions of the proposed analy-
sisby no means every unusual case is felt to be undecidable. But
nothing I have said implies that it should be. For that feeling to
arise the unusual features of the situation must engage with the every-
day practice so as to produce some contrary pressureto make
us feel that something that matters is lacking. Not every unfamiliar
circumstance need have that effect.
Certain worries call for mention at this stage, just in case they
are already beginning to worry the reader. First, it is not only per-
sons, but also for instance books and video-cassettes, that can be

sources of information; but although these are sometimes said to

contain knowledge, as is also a library, they are not said to know
anything. Why not, if the function of 'know' is to flag good sources
of true belief? One might have replied that it is because they don't
have beliefs, but what we have said in the earlier parts of this section
seems to imply that the present line of thought will not satisfactorily
explain why that should be a reason. So there is a problem here
which needs attention. Secondly, someone who has a strong motive
for concealing the truth may still be said to know it. But he is of
very little value as a source of information, except in those infrequent
situations where the inquirer is in a position to give him an even
stronger motive for revealing it. (Luigi knows exactly where Mario's
body is, and how it came to be there, but there is no point at all
in turning to him for information on the subject.) Thirdly, someone
who both knows the truth and is keen to reveal it may be useless
to others because he has no credibility with them: the boy who
cried 'Wolf!' so often that no-one would believe him when the wolf
really came is a cautionary example, as is Matilda of Hillaire Belloc's
Cautionary Tales. A gap threatens to open here between the con-
structed concept and the concept of knowledge as we operate it,
a gap wide enough to suggest that the former is too different from
the latter to throw any light on it. For the moment I would just
ask that we take note of these difficulties. Reacting to them involves
a complication of the picture that is best postponed, since the less
complicated version still has a good deal more to yield.
It is the third clause of the analysis, I need hardly say, which has
caused all the trouble. Is it that the belief has to be based on good
reasons, or that it has to have the right causal ancestry, or that
it must have been acquired by a reliable method, or that it must,
in Nozick's felicitous term, 'track' the fact that is its object? All
these, and probably more, are on the market, some in a number
of models. It has even been urged that we decline them all, and
define knowledge as true belief, with no adornments. Our next ques-
tion must be: does the concept that we are constructing, the one
whose sole purpose is to act as a marker for approved sources of
information, call for anything more than that, and if so, what?
First impressions suggest a short way with this question: no
further conditions are needed. Why should we want more of a poten-
tial informant than that his views on the point at issue should be
true, and at least confident enough for him to be prepared to come
out with them? Then we come to hear the truth, which was what
we wanted. But this overlooks a crucial point. It is not just that
we are looking for an informant who will tell us the truth about
p; we also have to be able to pick him out, distinguish him from
others to whom we would be less well advised to listen. How is
that to be done? Well, it will be easy enough to find out what
he believes about/;; and if we ourselves knew whether/? that would
suffice to tell us whether he has a true belief. But ex hypothesi we
do not know whether />we are in the position of inquirers, not
of examiners (to borrow Bernard Williams's way of putting it); the
informant is to be our means of access to that knowledge, and if
we already had it, we would not be inquiring. Obviously, we have
to detect the right informant without benefit of prior knowledge.
So we need some detectable property-which means detectable to
persons to whom it is not yet detectable whether/*which correlates
well with being right about p; a property, in other words, such
that if the informant possesses it he is (at least) very likely to have
SECT. Ill 19

a true belief on that matter. The emergence of this requirement gives

us, as it turns out, quite a lot to work with.
Let us at this stage set up a target for ourselves, so to speak.
Nozick1 has suggested that a knower must, besides having a true
belief, satisfy two further conditions:
(i) \{p were not true, he wouldn't believe that/?,
(ii) If p were true (but under circumstances differing slightly from
those actually obtaining) he would believe that p.

In discussing these conditions, I shall make free use of the now

popular vocabulary of possible worlds (without supposing it to be
anything more than a linguistic convenience); and I shall follow
David Lewis2 in saying that (i) and (ii), as counterfactual state-
ments, are true if and only if in possible worlds close to the actual
world, if p is false the subject does not believe that p, and if it
is true he does believe it. I shall ask whether the practical explication
we are engaged in might not lead to just these two conditions.
An immediate reaction is to ask 'Why should it?' Why should
our inquirer be interested in what is the case in possible worlds?
After all, he wants to be told the truth in this world, the actual
world, so whence the interest in other, and merely possible, worlds,
however 'close' they may be to this one? Either Nozick's 'tracking'
condition is not what we are looking for, or Lewis' semantics for
counterfactuals in terms of possible worlds have to be dropped.
But this is too quick: there is a line of thought which shows that
the inquirer cannot help being interested in the contents of possible
worlds as well as those of the actual.
We have to remember that the inquirer's knowledge of the actual
world is bound to be highly incomplete. It is not only that he doesn't
yet know whether p; there will be all sorts of things about himself,
the environment and the potential informant of which he is ignorant.
There are, in other words, enormously many propositions such that
he does not know whether A or not-^4, whether B or not-5, and
so on. So if we think of a world as defined by the totality of what
is true in it, there are indefinitely many different possible worlds
any one of which, so far as he knows, might be the actual world.
His concern with getting the right information in the actual world

1 2
R. Nozick, (2), pp. 172-8. See also F. Dretske. D. K. Lewis.
20 SECT. I l l

will therefore lead him to hope for an informant who will give him
the truth about/; whichever of all these possibilities is realised. Which
is to say, if you like the jargon, that he wants an informant who
will give him the right answer in a range of possible worlds.
Is it possible that we are not being quite rigorous enough here,
and claiming more than in fact follows from the inquirer's position?
The reply can be made that he is only interested in the actual world
it is just that he doesn't know, out of all those possible worlds,
which one it is. What I need, he might say, is for the informant
to be right in whichever of these possible worlds is the actual one;
the rest can go hangI shall have my true belief.
I doubt whether we can accept that, however. The trouble is
that it leads to no strategy on the part of the inquirer. Imagine
someone about to go out, and wishing to stay dry, but not knowing
whether it will be raining or not, so that he faces one 'possible
world' in which it will, and one in which it won't. Can he say
T only want to keep dry in the actual world; I'm not bothered
about whether I would have kept dry in whichever of those worlds
turns out to be merely possible'? He can say it, and in a sense no
doubt it is true. But if he proposes to do something about it he
will either have to guess which possibility will be realised or take
such action as will work in either case, even though that means
planning for at least one eventuality which will turn out to have
been merely possible. And the same applies to our inquirer: he must
either guess which of the possible worlds he is actually in, or he
must adopt a strategy which works in many merely possible worlds
as well as the actual one.
Some may now have got the impression that I am about to award
a prize to Professor Nozick; for haven't we said, in effect, that the
search for the desired informant will naturally become a search for
someone who is precisely a good 'tracker' of p, someone whose
belief as to whether p is true in all close possible worlds? But I
am not, for two main reasons. For one thing, Nozick does not
select the right range of possible worlds, if by that we mean one
that coincides with the range of interest of the inquirer. For another,
Nozick's tracking condition doesn't reflect the epistemic demands
that the inquirer is bound to make. I shall return to the second
point later, having dealt in some detail with the first.
The first point, then, is that Nozick's range of possible worlds
is not the one which will concern the inquirer; it overlaps with
SECT. I l l 21

it, of course, but it is wider. And just this extra width causes trouble
for his account, seen (in the way he surely meant it) as an attempt
in traditional style to match an intension to the intuitive extension
of 'know'. The practically explicated concept, on the other hand,
picks the right range of possible worlds, and illuminates the reason
for doing so. Here we see it scoring its first points against a presti-
gious attempt at an analysis of the more familiar kind.
Nozick recommends assessing his two counterfactuals by refer-
ence to what is the case in all close possible worlds. Fortunately
it is not necessary for our purposes to specify exactly what 'close'
means here. Roughly speaking, two worlds are said to be close to
each other if they differ only slightly, distant if they differ radically,
and this is accurate enough for the point I now wish to make, which
is that our inquirer will not be interested in all close possible worlds,
but only in those that he cannot rule out as being merely possible,
or non-actual. Suppose he is considering the credentials of a potential
informant whom he can see to be wearing a red shirt. There is a
close possible world (and it surely is close, if the concept of closeness
is to be capable of any work whatever) in which that same person
is wearing a blue shirt; but since the inquirer will be perfectly satisfied
that this world, although both possible and close, is not the actual
world, he will have no interest in it. It is of no concern to him
whether the potential informant would hold a true belief on the
question at issue were he wearing a blue shirt - he already knows
that he isn't.
(Here we should watch out for a point which could cause confu-
sion. If you really thought that I would quite likely be wrong about
p were I wearing a blue shirt, you would be doubtful about employ-
ing me as informant even if you could see that I am wearing a red
one. But that is not be because you are taking the 'blue shirt' world
into account as one in which, for your purposes, I need to be right.
Rather it happens because, your beliefs about the effect that the
colour of one's shirt has on capacity as an informant being what
they are, for almost any value of p the belief that I wouldn't be
good on p were I wearing a blue shirt will make you wonder how
then I can be good onp when wearing a red one.)
If we use the expression 'open possibility' to mean a possibility
which so far as the inquirer knows might be actual, we may put
the point more generally in these terms: the inquirer will want the
informant to be a good tracker of p in worlds that are both close
22 SECT. I l l

and open possibilities for him, the inquirer. This will be a narrower
class of possible worlds than that which Nozick uses for the assess-
ment of his counterfactual conditions. The possibility therefore arises
that Nozick's analysis will rule out some cases of (putative) know-
ledge where an analysis more closely tied to the situation of the
inquirer would allow them. This would happen if tracking held in
all the possible worlds which the inquirer will have an interest in,
but failed for at least one of the wider class of possible worlds which
Nozick would have us take into account.
With this thought in mind, let us look at a case which causes
Nozick a good deal of trouble: the case of the Great Bank Robbery. 3
Jesse James, the reader will recall, is riding away from the
scene of the crime with his scarf tied round his face just below the
eyes in the approved manner. The mask slips, and a bystander,
who has studied the 'wanted' posters, recognises him. The bystander
now knows, surely, that it was James who robbed the bank. But
Nozick has a problem: there is a possible world, and a 'close' one,
in which James' mask didn't slip, or didn't slip until he was already
past the bystander; and in that world the bystander wouldn't believe
that James robbed the bank, although it would still be true that
he did. So Nozick's condition (4)the second of the two counterfac-
tuals - is not satisfied, and he is threatened with having to say that
the bystander doesn't know that it was James, even though the
mask did slip. So his analysis looks like ruling out something which
is as good a case of knowledge as one could wish for.
In the face of this problem Nozick resorts to fudging. He recalls
his previous stipulation that the method by which the knowledge
is acquired be held constant, so that close possible worlds in which
a different method would be used to arrive at the belief that p from
that used in actuality are not to figure in the assessment of the coun-
terfactuals. And he then implies that, had James mask not slipped,
the bystander would have been employing a different method, an
implication which violates the distinction between the method and
the evidence obtained by the use of it (as Graeme Forbes4 has
put it). A natural response to the case of the Great Bank Robbery,
I believe, is that such a manoeuvre ought to be unnecessary; for
what would have happened if the mask had not slipped is wholly

3 4
' R. Nozick, (2), p. 193. G. Forbes, pp. 47-8.
SECT. I l l 23

irrelevant to the question of the bystander's knowledge. The right

approach to the matter would never have let it come anywhere near
the picturethen no fine and dubious distinctions would have been
necessary to expunge it again. But the view from the standpoint
of the inquirer has the properties we want here. Until he is satisfied
that the mask did indeed slip, the sheriff will not be interested in
the bystander's identification; once he is satisfied that it slipped,
possible worlds in which it did not slip are deletedthey play no
part, not even implicitly, in his judgement as to whether the by-
stander knows that it was James.
The second point about Nozick's analysis is this: What I have
said so far may suggest that the difference between his 'tracking'
and the conditions (in addition to true belief) which our inquirer
will want to impose on his sources of information is fairly minimal:
the inquirer will be happy with counterfactuals of Nozickian stamp,
assessed a la Lewishe will simply consider a narrower, more con-
text-dependent, range of possible worlds. If so I apologise, because
the suggestion is seriously misleading. Good tracking of the fact
that p cannot be the property that the inquirer is looking for; at
best it may be something that coincides with it.
I said earlier that Nozick's tracking condition does not reflect
the epistemic demands that the inquirer is bound to make. The
inquirer has to pick out the right informant, and no doubt it would
help, even if only as a first approximation, to pick out a good tracker
of the fact p. But he cannot set himself directly to pick out such
a person, because the truth of a counterfactual is not epistemically
primary in the sense that that would require. We don't have equip-
ment that allows us to spot people who satisfy Nozick's counterfac-
tuals as such. We can tell of some people that they do, and we
do it not directly but by noticing something about them, perhaps
the way in which they came to believe that p, which can be seen
to correlate with satisfying the counterfactuals. What we ought to
be looking for, at least in the first instance, is some such epistemically
more accessible property.

So long as we are cautious, it will do no harm to take another cue

from the current state of the literature and wonder whether Alvin
Goldman's causal theory 1 may not be what we are after. Could
the property be the causation of the belief that p by the fact that
/?? Certainly there can be few better guarantees of the truth of a
belief than that it was brought about by the fact in which it is a
belief. But in spite of that, or perhaps for that very reason, it is
not very well suited to the needs of our inquirer. Sometimes, of
course, we are in a position to say that a person's belief that p is
a causal consequence of the fact that/*; but under those circumstances
we already believe that/>, and do not feel the need for information.
On other occasions, admittedly, there may be things about a poten-
tial informant which suggest to us that his belief on the question
whether^ is caused by the facts: we saw him looking in the right
direction at the right time, for instance. And if it suggests that,
then a fortiori it suggests that his belief is true. But at the same
time the importance to the inquirer of the existence of the causal
connection is diminished. For his experience is that most beliefs
about the location of the cat that are formed by someone who is
looking hard at the mat turn out to be true; and this will lead him
to place confidence in their opinion, whether or not he believes
that seeing is a causal process. Human beings were good judges
of a likely source of information long before that sort of question
was ever thought of.
So there is no particular reason to think that the concept to which
the inquirer's perspective leads us will have to contain a clause about
causal connections. Like tracking, causal connections with the fact
believed are likely to be frequent correlates of the property X, but
that is all. As to -what the property X may actually be, we are still
in the dark.
Points arising out of our discussion of the analyses offered by
A. Goldman, (1).

Nozick and Goldman do, however, indicate an answer to that ques-

tion. It is no very precise answer, but perhaps the whole point
is that the property X has no very precise identity. What it suggests
is simply this: X is any detectable property which has been found
to correlate closely with holding a true belief as to whether p. (Some
readers will think that we ought not to speak here of the particular
proposition p, but rather of 'propositions of the same type as p'.
They have a point, and I shall take it up later in the slightly different
context of Section VII.) Any such property will give the inquirer
what he needs, so long as we add one proviso: that the correlation
be lawlikean accidental correlation, one that does not support
inference to a new case, will not do, since inference to the new,
as yet untested case is precisely what the inquirer needs the correla-
tion for.
In so far as such correlations are going to be causal, this lends
some substance to the thought that there must be a causal connection
between possessing the property X and being right on the question
in hand. But this is not the same as saying that the belief in p must
be a causal product of the fact that pit does not give us the causal
analysis as that is normally understood. It also chimes in with the
idea that counterfactuals are in the offingthe ability to support
counterfactuals has often been said to be a defining characteristic
of such lawlike connections. But notice that these aren't Nozick's
counterfactuals, which link p/not-p counterfactually to the belief/
absence of the belief that/>; rather these link the counterf actual pos-
session of property X to the possession of the truth about/?. (They
say, for instance: if Fred had been looking, he would have been
right about where the cat was.)
A natural way of describing the position reached would be to
say that the correlation between possessing X and being right about
p must be reliable. When he encounters A* in a new case, his inference
that its possessor will have the truth as to whether p must not lead
the inquirer astray. If it can be reliable without being causal (I suspect
it can), then being causal is a special case of the general principle;
if it can be reliable without supporting counterfactuals (though this
time I suspect it can't), then that is a special case too. But notice,
once again, that this is not the same as saying that the property
is that of having arrived at the belief by a reliable method, in the
sense of one which (nearly) always leads to true beliefs. That property
does, for obvious reasons, correlate reliably with being right; but

whether it is the only property that does so is quite another question.

Prima facie it seems hardly likely. After all, it might have been the
case, had we been rather more scrupulous about our testimony and
rather less prone to error, that the mere fact of being willing to
offer an opinion correlated excellently with being right (indeed, as
I said earlier, for some classes of beliefs this is pretty much the
case as things are); and nobody, surely, will be prepared to stretch
the concept so far that being willing to offer an opinion can be
called a method of acquiring belief. The idea that there is such a
correlation may be exaggerated but, as I say, it is not wholly ficti-
tious; and at the very least it shows that it is a live question whether
there are any actual states that are reliable indicators of true belief
without counting by any distortion of the vocabulary as the state
of having employed a certain method.
There surely are such states. If you want to know the way it
will always be a good idea to ask a taxi-driver. The advice is good,
because the property (which has the additional advantage of being
fairly easily detected) of being a taxi-driver correlates very well with
giving the right answer to that kind of question. And the property
of being a taxi-driver certainly can't be identified with that of having
discovered, by a reliable method, how to find one's way round
the neighbourhood, although of course it is true of taxi-drivers that
they have found that out by a reliable method: driving round the
place for hours and hours day after day. But our inquirer doesn't
need, strictly speaking, to impose any conditions on how they do
it; so long as he believes that they do, and can recognise them,
he can acquire the best information. It would be just the same if
taxi-drivers were born that way, and never had to do anything what-
ever to come by their grasp of local geography.
The point can be generalised. A very large part of the art of acquir-
ing correct information consists in being able to recognise the sort
of person (or book, or whatever) that wdl have the right answer.
So long as the inquirer is right in thinking that it has it, he will
not have to concern himself with how it may have come by it,
whether by a reliable method, or as a causal consequence of the
relevant facts. So no such clause will appear in the concept which
we are constructingonly the condition that the subject possess
some detectable property that is a good indicator of true belief on
the matter under discussion.
It will by now be seen that there is not going to be any more

detailed answer to the query: and what property is that? There could
be almost as many different answers as there are types of thing that
the inquirer might want to know about. If it were not so, choosing
our sources of information would be a very much simpler business
than it is. But even the very general formulation that we can give
is enough to enable us to link the constructed concept to the most
prominent recent attempts at an analysis of the concept of know-
ledge. Let us begin by looking back to the counterfactual analysis
of Nozick and Dretske. It adds as the further condition the two

(i) not-p not-(S believes that/;)

(ii) S> believes that/?

Now in the preceding section I have in effect argued three things.

First, that the inquirer must indeed be looking for an informant
whose belief as to whether/7 satisfies such counterfactuals; that arises
from the human inevitability that there are many possible worlds
any of which, so far as the inquirer either knows or even believes,
could be the actual world. Secondly, but in this context unimpor-
tantly, that Nozick sets the range of possible worlds across which
(i) and (ii) are to be assessed too wide.
Thirdly I have also argued that, for epistemic reasons, (i) and
(ii) cannot be all that the inquirer is looking for. But here the word
'all' is crucial. The epistemic point does not affect the fact that the
inquirer needs an informant of whom (i) and (ii) holdit only shows
that he needs something else as well, namely something to assure
him that they hold.
We can therefore see why, if my hypothesis be correct that the
core of our conceptual practice using 'know' is what is constructible
by taking the perspective of the inquirer, an analysis such as that
of Nozick and Dretske should be felt to be close to the mark. We
have not, it must be admitted, seen why it should be felt plausible
to finish with the conditionals (i) and (ii), since the inquirer needs
something further by which to detect the informantwhich the truth
of the counterfactuals by itself cannot enable him to do. But we
can quickly convince ourselves (I shall consider this issue more
slowly in Section X) that it would be easy to argue for leaving that
further element out of the analysis. The informant knows, it could
be said, so long as he meets the traditional requirement of true belief
28 S E C T . IV

plus the two 'tracking' counterfactuals. The inquirer, however, needs

to be able to tell that he knows, so he must exhibit some 'indicator
property' X as well; but we are not obliged to take that additional
X up into the analysis of knowledge.
Once we can see why the tracking analysis should appeal it is
no great step to do the same for the Causal Theory. Of Hume's
various 'definitions' of cause one, we may remember, was that 'if
the first object had not been, the second never had existed';2 and
following that line of thought we would be lead to identify the
Nozick-Dretske counterfactuals with the claim that the subject's
belief that/) be caused by the fact that/?. Evidently these two analyses
are no great distance apart.
Although the distance isn't great they are not, of course, identical.
For that there are too many problems of detail about Hume's coun-
terfactual definition of causation. But it is illuminating to think,
bearing the literature on the concept of knowledge in mind, what
some of these problems may be. One is that it makes certain processes
automatically causal in nature when one would have thought that
that question was at least open to discussion. If, for instance, I
believe that A, infer from A that B and so come to believe that
B, then the two states of my believing A and my believing B will
stand to one another in the critical counterfactual relationship. Are
they cause and effect? Only if inference is a causal process. Perhaps
it is, but should this really be decidable in one line ex definitione}
There is a hitch here, but, significantly, it is no problem for Gold-
man's causal analysis; for Goldman simply stipulated that for the
purposes of his 'causal' theory inferential processes were to count
as causal.3 Given this stipulation, the causal condition moves even
closer to the third condition of Nozick and Dretske.
Goldman also said quite explicitly4 that his analysis was aimed
only at a posteriori knowledge; analysing 'know' as it applies to
a priori matters was not on his agenda. With this restriction the
causal condition edges still nearer to the counterfactuals. For belief
in a priori truths was one of the main things holding them apart.
Consider my belief that the differential of x" is nx"~l. Were it not
so, I would not believe it, since I would not have seen a proof
of it, nor would my teachers have told me that it was true. And

2 3 4
D.Hume, p. 76. A. Goldman, (1), p. 362. Ibid., p. 357.

given that it is so, quite a lot of (small) changes could have been
made to the world as it is without dislodging my belief. So the
truth of the theorem and my belief that it is true satisfy the two
counterfactuals. But to say that they are causally related embroils
us in all manner of ontological difficulties which surely go far beyond
anything we have said in merely agreeing that the counterfactuals
hold: if the truth of a mathematical formula can be the cause of
a belief must it not designate some kind of state of affairs, something
of the sort which could enter into causal relations? And then what
sort of state-of-affairs could it be, and how could its influence on
our minds be a causal one? This is well-trampled ground, but fortu-
nately there is no need for me to trample it still more. My point
is simply that removing all this from consideration, as Goldman
did in his early paper on the Causal Theory, helps to bring causality
and tracking into near-congruence by excluding a principal area in
which they may very plausibly be thought to diverge.
The next step is to see how close we now are to another popular
formula, that which declares knowledge to be true belief attained
by a reliable method. For some writers, indeed, there seems to be
no distance at all between this formula and one or other of those
we have looked at already. Ramsey5 once defined knowledge in
terms of the 'reliable method' formula, but then quickly said that
what he meant by a reliable method was, or at any rate included,
causal connection with the fact believed in. More recently, David
Armstrong6 made similar opening moves, and then characterised
reliability in terms of satisfaction of the Nozick-Dretske tracking
conditionals. But even without adopting so direct or stipulative a
stance we can easily establish contact between the account in terms
of reliability of method and those in terms of tracking and causal
One's first thought might be that if a certain belief satisfies the
tracking conditionals, and if in addition there is such a thing as
the method by which it was acquired, then that method must be
a reliable method, in the sense of one which will produce the belief
that p if p is the case, and will not produce it otherwise. But the
first thought needs a supplement: it must be that the tracking con-
ditionals are satisfied because the belief was acquired in that way.

1 6
F. P. Ramsey. D. M. Armstrong, ch. 12, esp. p. 169.

Were that not so, then although the belief might be called reliable,
there would be nothing particularly reliable about the method.
Now for the vast majority of human beliefs, if not indeed for
all, there is such a thing as the method of acquisition. We do not
just have beliefs, we come by them in specific ways. Perhaps in
creatures whose mental life is much more primitive and much more
instinctual than ours there are beliefs which they have not in any
real sense acquired; perhaps even some very few human beliefs are
like that too. But be that as it may, the vast majority of human
beliefs are acquired and something, at least, can be said in description
of the process by which their subject acquired them.
It does not strictly follow from this that when we trust a belief
we trust it because it was acquired by such and such a methodone
with a long and successful career behind it. But although that does
not strictly follow it is nevertheless a fact that what convinces us,
in an enormous number of cases, to place confidence in a certain
belief, is a further belief about the way in which its ownermaybe
another, maybe ourselfcame to hold it. And even in those cases
in which what did the trick was better describable as something
about what the belief-holder was than as how he came to believe
ita teacher, perhaps, or a policeman, or a doctor-very often some
view about how such people come to hold their beliefs on the appro-
priate subject-matter underlies and underpins our confidence in their
opinions. For all such cases, therefore, the idea that what we are
looking for is true belief reliably acquired will fit the bill; and since
they form so large a proportion of all the cases we are likely to
consider it should not be felt surprising if that formula has seemed
to offer an acceptable definition of knowledge in general. For in
virtually every case in which we take the counterfactual conditions
to be met, we also take it that the reason why they are met lies
in the nature of the method used.
It seems then that where there is tracking in consequence of the
method used we have a reliable method. But the relationship between
reliabilism and the tracking and causal analyses would fall apart if
the converse failed, in other words if a method could be reliable
without producing beliefs that satisfied the counterfactuals. That
would be so if a method which generated true beliefs just in all
actual cases could thereby qualify as reliable. And so long as we
do not take into account the point of the concept of knowledge
we might think that reliability could be so defined. But a reliable

method, if it is to have any role to play in the construction of the

concept of a good informant, must be a method one can actually
rely onand in Section III we have already seen how this demand
brings the counterfactual property in its train.
In spite of the fact that it was alive and well in the minds of
most philosophers only thirty years ago, we have not yet paid any
attention to the traditional account of knowledge in terms of true
belief with a reason or justification. But it should not be thought
that only its recent successors are capable of the kind of treatment
we have just been discussing. There are good grounds for thinking
that where the minimal concept of the good informant applies, there,
very nearly always, we will find true belief with a good reason as
well, provided only that the notion of having a reason for a belief
is not taken too strictly. Let us say that 5 has a reason for his belief
that p if there is something else, q, such that:
(1) It is true
(2) S believes it
(3) Its truth significantly raises the likelihood thatp is true
(4) S believes that (3) holds
(In including (4) I may appear to support an 'internalist' account
of what it is to have a reason; but I do not wish to argue with
those who would have us stop after (3). My purpose is merely tacti-
cal: since the argument I shall now give goes through if we include
(4), andas will be seengoes through even more easily if we do
not, it seems sensible to put (4) in and satisfy both the internalist
and the externalist at once.)
Now it is usually true of human beings that they are prepared
to offer something by way of a reason in support of any belief they
hold. And 'usually' rises to 'nearly always' when we recall that (for
reasons given in Section II) we will nearly always be dealing with
beliefs that are held with some confidencecases like that of Colin
Radford's French Canadian will be very much the exception.
Further, we are not just speaking of beliefs, but of beliefs in which
5 really is right and in addition to being right possesses some property
which reliably goes with being right. We have seen that such beliefs
are generally acquired by certain standard types of process, and
human beings are normally aware of some of the more salient stages
of these processes. ('Salient', admittedly, doesn't mean anything
more than that they are the ones that we tend to be most aware
32 S E C T . IV

ofbut the point that follows doesn't need it to mean any more
than that.)
In the case of a (true) belief acquired perceptually, for example,
S will normally be aware of having had such and such visual (or
other) experiences. It will be true that he has indeed had them,
he will believe that he has had them, that he has had them will
raise the likelihood that his belief about the state of his environment
is a true one, and he will believe that it raises the likelihood. So
the above conditions (l)-(4) will be met. They will also be met
if we consider the standard cases of coming by beliefs about math-
ematics. There will be some proposition q to the effect that 5 has
been through a process of proof or calculation that terminated in
p, S will believe it, its truth will raise the likelihood that p is true,
and S will believe that as well. The position is the same if we consider
inferentially acquired beliefs of an a posteriori kind. Suppose I
believe, truly and with some confidence, that there has been a dog
in the garden. I will normally have come to this belief via the belief
that there are paw-prints in the flower-bed or something similar.
This something, whatever it is, will nearly always be true, since
the occasions on which I will come to a true belief as the result
of a false belief will be very much in the minority. The first two
conditions are, therefore, satisfied. So are the others, for the truth
of q (in this example: that there is a paw-print) significantly raises
the likelihood of p and I will virtually always believe it to do so.
Again, suppose that I believe that there is a dog in the garden,
this time because I see it. Whether we describe my reason (the propo-
sition q of conditions (l)-(4) above) as 'my having seen the dog'
or rather as 'my having had a visual experience as of a dog' is a
question we can postponeindefinitely, if we so wish. It remains
true that (1) some such thing has happened, (2) in very nearly all
cases I will be aware that it has happened, (3) its having happened
increases the likelihood that there is a dog in the garden, and finally
(4) I will be aware that it increases the likelihood that there is a
dog in the garden.
If, then, 'having a reason' is captured by (l}-(4), normal human
consciousness (and self-consciousness) of the processes leading to
their beliefs will ensure that they have reasons for nearly all those
beliefs with respect to which they are reliable informantsor which
they 'know' by the lights of the constructed concept; a fortiori i the
same holds if 'having a reason' is sufficiently covered by (l)-(3).

Like true belief that tracks the facts, or true belief causally connected
to the facts, or true belief attained by a reliable method, true belief
with a reason is not at all far off the mark. Once again it becomes
understandable that philosophers, especially philosophers interested
in emphasising the role of rationality in human life, should have
presented the concept of knowledge in this dress.
It may be worth pausing at this stage to draw a certain parallel.
There is a similarity between some of these thoughts and a proposal
once made by H. P. Grice7 in connection with another question.
Grice was roasting one of the old chestnuts of philosophical logic,
the relationship between 'If. . . then. . .' as it occurs in everyday
speech, and the material implication (p*q) of formal truth-
functional logic. Can one equate them? There are, of course, excel-
lent and over-rehearsed reasons for thinking that one cannot: the
material implication comes out true, for instance, if any pair of pro-
positions, however remote and unconnected, are substituted for p
and q, provided just that the first of them is false, or both are false,
or both true; whereas 'If I am writing about Grice, then the Moon
is about 240,000 miles from the Earth' is surely not trueat any
rate it is extremely weird and would never be said. But if they meant
the same then the second would have to be true if the first wasso
they do not mean the same. Grice undertook to show that this
well-worn argument was indecisive. He formulated certain general
principles allegedly governing all conversation, and so having
nothing in particular to do with the meaning of 'If.. . then.. .'. He
then proposed that an everyday conditional, 'If ... then .. .', means
no more than the corresponding material conditional, and that those
features of its use which suggest otherwise do not result from its
meaning but from the operation of these general principles.
My procedure with respect to the concept of knowledge is to
some extent analogous. I allow the situation and needs of the inquirer
to generate a concept, or, as it might be better to say, a description
of a prototypical instance, roughly: 'true belief plus some property
indicative of true belief, and then suggest, in effect, that we take
this to be the core of the concept of knowledge, as Grice suggested
that we regard the material implication as the core meaning of the
expression 'If ... then . . .'. We are then confronted with a number

H. P. Grice. (Note that this reference offers mainly material about the general
principles governing conversation, not about the specific issue of conditionals.)

of analyses, none of them, evidently, very far from the mark; and
I make no attempt to deny that there is some justification for the
additions they each make to the minimal concept, much as Grice
accepted that the actual use of the everyday conditional does differ
from that which formal logic lays down for material implication,
and that an attempt to reflect these differences in a different account
of its meaning consequently has some foundation. But these very
genuine differences, Grice thought, were not best described as a
difference in the meaning of the expressions; and in much the same
way I suggest that the differences between the 'minimal' analysis
and the competing 'standard' analyses are not best seen as the out-
come of the concept of knowledge. What makes them seem plausible
is not the concept of knowledge, but certain very general beliefs
which we all hold. These are, in particular, beliefs about the extent
to which the world is a system of causally inter-related states, more
specifically beliefs about the extent to which belief-states are them-
selves the end-product of a causal process; the belief that for nearly
all human beliefs, there is such a thing as the method by which
they were acquired; and the fact that human beings are usually con-
scious of certain stages of the processes by which they arrive at
beliefs. The effect is that when the conditions laid down in the mini-
mal concept are satisfied, it will almost always be believed that vari-
ous further conditions are satisfied too: in particular, those of Nozick
and Dretske, those of Goldman's causal theory, those of the reliabi-
list in so far as they genuinely differ from these; and finally (to
reverse the chronology) those of the traditional analysis of know-
ledge as true belief with a reason.

We now need to pick up a hint from the last section, and take
a closer look at a distinction which I have so far smudged over,
though it is an important one. There are informants, and there are
sources of information. Or, to arrange the terminology differently,
among the various sources of information there are on the one hand
informants who give information; and on the other there are states
of affairs, some of which involve states of human beings and their
behaviour, which have evidential value: information can be gleaned
from them. Roughly, the distinction is that between a person's telling
me something and my being able to tell something from observation
of him. Of course, in this evidential sense it is far from being only
persons that are sources of information. A tree is a source of infor-
mation on its age, since one can tell its age by counting the growth-
rings ; in fact, anything is a source of information on a great variety
of mattersgiven a suitably equipped observer who knows which
inferences to draw.
In general terms it can be said that the concept of knowledge,
as we operate it in everyday practice, is tied to informants rather
than to sources of information in the sense just (approximately) char-
acterised. We don't speak, even metaphorically, of a tree as knowing
how old it is; and if Fred enters dripping wet, although he may
well know that it is raining, we don't say that he knows it just
because we can tell it by looking at him.
Just to draw this distinction isn't enough; we need some idea
as to why it should have been felt worth drawing, otherwise we
shall find that our 'practical explication' leaves us still a very long
way from the concept of knowledge as we find it in use. Why bother
to build into the vocabulary a distinction, amongst sources of infor-
mation, between 'informants' who 'know' on the one hand and other
sources which do not know on the other? The answer cannot lie
in the distinction between agents and non-agents, for the second
example of the previous paragraph proves that an agent can be a
36 SECT. V

source of information in both ways; indeed an agent's actions can

be sources of information in both ways, as a little further thought
shows. And even if this were not so, we would still be left wondering
why, if what we are after is true belief, we should attach importance
to the distinction between getting it from something that acts and
getting it from something that doesn't. More promising are two
other thoughts, namely (i) the convenience of an informant as
opposed to a 'mere' source of information, and (ii) the special psy-
chology of team-work in a community, something which is involved
in the use of informants but not in the use of sources of information.
The point about convenience is that anyone who understands both
the question (a test which the inquirer can hardly fail) and the lan-
guage he speaks or gestures he uses can get information from an
informant, whereas getting it from a 'source of information' "will
often call for various degrees of more specialised knowledge, perhaps
for a special ability to evaluate evidence. In comparison with (i),
(11) is more questionable, and far harder to pin down; for this reason
it would be awkward if one had to restas we fortunately do not
the whole case for the importance of the informant/source of infor-
mation distinction upon it. But it is still worth mentioning. What
I have in mind is the special flavour of situations in which human
beings treat each other as subjects with a common purpose, rather
than as objects from which services, in this case true belief, can
be extracted. 1
It isn't all a matter of 'special flavour', a little additional cosiness.
There is another, and this time much less elusive, factor: an informant
is a co-operating member of our species. That means that he can
often empathise with the inquirer, and react not just to the question
but to the presumed purpose of asking it, so giving the inquirer
useful information that he didn't know he had need of. 'Where's
the bus stop', he asks, and is told not just 'Fifty yards down there
on the right', but in addition: 'And that's the last bus just turning
into the road.' Mere sources of information, on the other hand,
though they may often be extremely useful, are never actively help-
ful. How could they be? They don't know what the inquirer is
up to.
Once we have this distinction in view a number of other lines
of thought open up. I should like to begin with this one: that if

' In this connection see A. Ross.

SECT. V 37

knowing has to do with being an informant as opposed to just being

a source of information, this could do something to explain our
reluctance (which does not seem to amount to complete refusal)
to regard someone who does not believe that p as knowing that
Let us begin with a case in which we can use as a source of infor-
mation as to whether p someone who would certainly not be said
to know whether p. Fred is, as I am aware, systematically wrong
about what day of the week it is: he is always a day behind. Now
I can certainly find out which day it is by asking him: if he says
Thursday, I can rely on its being Friday, and so on. But Fred neither
knows what day it is, nor is he a good informant; for he does not
tell us, or even believe, that it is Friday. Likewise, if there are two
identical twins well known to me, Judy and Trudy, and Fred knows
one but is not even aware of the other's existence, then I may find
out that Judy is in town from Fred's telling me that he has just
seen Trudyfor I happen to know that Trudy is somewhere else,
so it must have been Judy that Fred saw.
Now we may move to the case in which the 'informant'if rightly
so calledactually tells us the truth: Radford's French Canadian,
Jean. Jean comes out with the right answers to questions about
British history, and someone who knows the background to this
surprising fact can certainly find out about British history by accept-
ing what Jean says. Notice that he does accept what he says; it
is not that he infers that p from Jean's assertion that q, as in the
examples of the previous paragraph. Nevertheless, that does not
feel wholly to settle the question whether he is treating Jean as
informant, or only as source of information. One can understand
the hesitation. The situation is extremely like the familiar one in
which we ask a straight question and receive a straight and reliably
correct answer. 'Tell me', says the interrogator, 'when did Elizabeth
I die?' And Jean says '1603'. But something is missing: does Jean
actually tell us these things, or does he just come out with numbers
in response to our questions? Or something lying mistily half way
between? So Radford's example does not (for reasons further to
those mentioned in Section II) absolutely force acceptance of the
view that there can be knowledge without belief. For an opponent
could say that where there is no belief there is no genuine communi-
cation or 'telling', and argue that we should therefore assimilate
Jean's case to one in which a piece of behaviour serves as an excellent
38 SECT. V

source of information (that is, a very useful piece of evidence) to

any audience with the needful background knowledge. For this view
of the matter the fact that the behavioural evidence in question hap-
pens to be a verbal expression of the right answer is inessential and
Radford prefers a different assimilation: crucial for him is the
undoubted resemblance to the standard and familiar situation.
'Prefers' may well be the operative word. Neither side has said any-
thing false. Both similarities exist, both are striking. And they are
not quantifiable, so is it any wonder if we are torn between them?
These thoughts can also throw light on certain cases mentioned
earlier2 as prima facie difficulties for our basic thesis: books and
the like, excellent sources of information, but never, even in the
spirit of metaphor, said to know anything. In terms of the distinction
we are considering, they are regarded not as informants but as
sources of information. Not that specialist knowledge of any kind
is required to unravel their secretsa large part of their point is
to provide a perspicuous source, accessible to anyone with a com-
mand of the language they use. But they have none of the psychology
of the prototypical informant: they have no beliefs, they do not
act, they are not felt to co-operate with us, and they cannot empathise
with us so as to anticipate our purposes. Besides, they have a special
place amongst sources of information: they are the evidence laid
down by creatures that are prototypical informants precisely as the
most perspicuous vehicle of their information. The most natural
thing for us to assimilate them to is the voice of a speaker, something
which, if I may be allowed a modest phenomenological flourish,
we 'hear straight through' to the person speaking. I would not deny
that it is possible, in some contexts even legitimate, to think of
them as some kind of evidence; but that is a highly unnatural posture
of mind which we can achieve only with sophisticated theoretical
The distinction between informant and source of information also
has a role to play in the discussion of another area of the topic;
and it will turn out later that its effect in this area is important.
Some writers have drawn attention to what they call a 'comparative'
aspect of the concept of knowledge: a person might be said to know
that a piece of music was by Mozart, in circumstances where the

Above, Section II, last para.
SECT. V 39

alternative was that it was by Bartok, but not to know if circum-

stances were such that the alternatives included Hadyn. Such exam-
ples, if genuine, cannot help but cause difficulties for an account
of knowledge which sees it as a relationship between the subject
and the fact that the piece is by Mozart, for it must be explained
how this relationship comes to be altered by the circumstance that
if it were not by Mozart, it would be by Bartok (those being the
only two pieces on the radio at the moment). But if we think of
the inquirer looking for a good source of information, we see that
these 'comparative' considerations can easily find a place. For the
inquirer may already have reached a stage at which the alternatives,
for him, have been reduced to just those two (for he has checked
with the published programme). If he has, then his standards for
an informant will naturally be much less demanding than if he has
not. (And someone who has not already excluded the possibility
that all our experience could be the work of Descartes' demon might
set his standards impossibly high.)
The inquirer-based conception therefore seems to fit comparati-
vism like a glove. One might be inclined to think that it fits it no
better than does the tracking analysis, which is of course perfectly
placed to bring in considerations about what would have happened
had the piece not been by Mozart. That may be, but how well
placed is it to bring in the point, vital to the example, that if the
music had not been by Mozart it would have been by Bartok? What
is so 'close' about the world in which Bartok was to be heard?
Granted that the world in which the radio is tuned to the other
channel isn't far away, but neither, so far as we can see, is a world
in which the channel that was in fact playing Mozart is playing
Haydnperhaps the compiler of programmes tossed a coin, heads
Mozart tails Haydn. And if that world were adjudged close,
the claim that the subject knows that the piece was by Mozart
might well fall, whereas it would stand if we considered only the
Bartok-world. We can see far more readily, I believe, why the
Bartok-possibility takes preference over the Haydn if we allow these
preferences to flow from the state of mind of an inquirer who has
already satisfied himself that Mozart and Bartok (but not Haydn)
are the relevant alternatives.
Unfortunately, however, there is a strong suspicion that compara-
tivism is spurious, and that the concept of knowledge presents no
such phenomena as those we are trying to explain. At best, its status
40 SECT. V

is much like that of the beliefless knowledge of Radford's Jean, as

I shall now argue. Let us first distinguish two situations:
(A) The inquirer, knowing that whatever was on the radio must
either have been Mozart or Bartok, selects an informant who will
give the right answer when the question is put to him as a disjunction
of those two possibilities.
(B) The inquirer puts the question in the open form ('Which
composer wrote the music you were listening to this morning?'),
receives the answer 'Mozart' and takes that to be the right answer.
Here again we need to distinguish two possibilities:
(i) He believes his informant to be very good at distinguishing
Mozart from any other composer.
(ii) He believes that his informant would at any rate not say that
it was by Mozart had it been by Bartok. (Though might had it
been Haydn, or early Beethoven, or C. P. E. Bach.)
Now surely there is at most one of these cases in which we should
feel decisively (given a moment's thought) that the informant knew
that the piece was by Mozart, and that is (B)(i). For on reflection
we immediately see that nothing is needed for (A) except that the
informant be able to tell, of anything that is de facto by Mozart,
that it was not written by Bartok; strictly speaking no more than
that is needed for (B)(ii) either, though one would have to accept
an element of coincidence, in that the informant has correctly picked
Mozart out of the bunch of composers whom he might just as well
have chosen on hearing that music. What is happening here, m fact,
is that the inquirer is taking the informant's utterance, putting it
together with what he (the inquirer) knows about the situation
(including the capacities of the informant), and deducing the truth
on the question he is interested in. He is using the so-called inform-
ant's utterance as a piece of evidence, not as a piece of information
this is obscured by the fact that the same words in the mouth of
a different person (the informant of (B)(i), for instance) would be
used as information.
Notable about (B)(i), of course, is the fact that the inquirer's special
knowledge is playing no part. It might have done so if the informant
had said that the piece was by Bartokfor the specification of case
(B)(i) does not guarantee that this informant can tell Bartok's music
from that of every other composer, only that he can tell it from
Mozart's. And if there is a variety of other composers whom he
cannot reliably tell from Bartok we would resist the idea that he
SECT. V 41

knew that this music was by Bartok even though he got the answer
right, and at the same time enabled the inquirer to conclude confi-
dently that it was indeed Bartok he had been listening to. The ascrip-
tion of knowledge, in other words, seems to stop just where specific
knowledge on the part of the inquirer is needed if he is to make
use of the 'informant'; and this coincides with the distinction we
intuitively make between using someone as an informant and using
their behaviour (including their utterances) as a source of infor-
mation, or more precisely as evidence. Notice that in case B(ii) our
inquirer would have got the information he wanted (that the music
was by Mozart) had he been told that the music was by Haydnso
long as he believed that the 'informant' would not at any rate take
Bartok's music for Haydn's. Once we have the crucial distinction
in mind we see that the informant is not clearly functioning as an
informant at all.
Ascriptions of knowledge may, as we shall later see, be in a certain
sense relativised to an inquirer's concerns. But they are not relati-
vised to any special knowledge of the circumstances surrounding
his inquiry.
Alvin Goldman3 has drawn attention to facts about the phenome-
nology of 'know' which are surely germane here. Henry, driving
in the countryside, sees a barn, and we may imagine circumstances
to be such that we unhesitatingly agree that he knows that that
is what it is. (Circumstances are, in other words, pretty much nor-
mal; and nobody has recently been putting any of the sceptic's
favourite possibilities into our heads.) Goldman now asks us to sup-
pose that we hear that, unknown to Henry, this stretch of country
is full of papier mache barn-facades, very plausible ones, which
Henry would not be able to distinguish from a real barn given the
amount of attention he can spare from the driving seat. Now do
we think he knows that this barn (this one really is) is a barn?
Most of us will think not, in spite of his being, in fact, right in
this case. Had he been 2 miles further down the road, let us say,
he would have looked out of the window, unhesitatingly identified
a barn, and been wrong.
Now vary the conditions. Suppose that there is only one such
fake barn, but that it is very close to the barn that Henry has (cor-
rectly) identifiedperhaps the very next one down the road is the
A. Goldman, (2), esp. Section I.
42 SECT. V

one that will catch him out. Decide how you feel, then try another
variation: there is only one such fake barn, and it is nowhere near
Henry's route. Or there was once such a 'barn', but it was destroyed
5 years ago. Or there never was such a barn, but someone once
thought of making one and then decided not to bother. Or it is
just that it is a physical possibility to make one.
My own reaction here (and I take it that it will be widely shared,
or that I will quickly be notified if it isn't) is that at the beginning
of the sequence I am reluctant to say that Henry knows that he
is seeing a barn, but that thereafter my reluctance falls off sharply.
How does this compare with the reactions of an inquirer weighing
Henry up as a possible informant?
We need to be very careful here as to exactly what the inquirer
knows at the outset. If he just knows that the fake barn is very
close to Henry, and here is Henry now announcing a barn, then
obviously he will be unwilling to use Henry as an informant, much
as we are unwilling to allow that he knows. But then comes an
anomaly. If the inquirer knows all that and a little bit more, namely
that the fake barn is not, as a matter of fact, the thing that Henry
is now looking at, won't he then be happy to rely on Henry to
tell him whether or not the thing he is looking at is a barn? Surely
he will, so long as he can eliminate the possibility that it is the
fake barneverything else around here Henry can perfectly well
distinguish from a barn. Yet the knowledge that it isn't the fake
barn that Henry is looking at doesn't increase our willingness to
say that he knows that the thing he is looking at really is a barn.
There seems to be a mismatch here between our intuitive use of
the concept of knowledge and the inquirer's view of the adequacy
of an informantso the discrepancy has to be explained.
We can explain it by calling once again on the distinction between
the informer and the source of information. Henry ceased to be
a desirable informant when we heard that there was, close to him,
a fake barn-facade which he would not manage to distinguish from
a real one. Receiving the further news that whatever Henry was
looking at when he announced a barn was at any rate not the said
barn-facade, and being satisfied that there was nothing else that he
would mistake for a barn, we could make use of him after all. But
notice that 'make use of seems to be about the right way of putting
it: it is only because of some special piece of knowledge which our
imaginary inquirer happens to have (and Henry doesn't) that he
SECT. V 43

can take Henry's affirmation as a basis for the belief that there is
a barn there. Because what he comes to believe is identical with
what Henry told him there is some temptation to think that Henry
is here informing him of the barn. But in view of the way in which
the inquirer uses his special ancilliary information, and wouldn't
form a belief without it, the case is just as easily assimilated to that
in which Henry announces a barn and the inquirer, aware that at
any rate there is no barn there, infers that it was the barn-facade.
Tentatively, we may make a more general point. Clear is at any
rate this: the fact that whether the object he sees were a barn or
a barn-facade Henry would say that he saw a barn, is something
that will not in practice not bother our inquirer. More accurately:
it will not bother him unless he assigns a significantly high prob-
ability to the alternative that it is a facade. This, I conjecture, is
what is at work on us in the series of situations just envisaged.
The mere logical possibility of a indistinguishable object gives no
boost to the thought that it is the fake that Henry is looking at,
the physical possibility (by itself) doesn't do so either; even the
knowledge that someone once conceived the intention of making
one has very little effect. If we are told that they actually did it,
then something stirs; and if we are told that they did it near here
the matter begins to get serious. If, finally, we hear that there are
several of the things along this very stretch of road, then we cease
to look to Henry to tell us where the (true) barns are. And our
judgement as to whether Henry knows that it is a barn (given that
in fact it is) tends to keep step with this progression: at the beginning
he does, at the end he doesn't, in between we waver in varying
Now, in one of the cases in which the likelihood that this is the
barn-facade seems to have grown large enough to matter, imagine
our inquirer to be in possession of additional information which
lowers that likelihood again. Our trouble was at first that he may
still use Henry as an informant, but we then saw that this is not
really what he is doing: he is using Henry in conjunction with his
own additional information to arrive at a belief, and so Henry is
serving him as a source of information. When we asked earlier what
gave us a practical interest in the distinction between informant and
source of information, a central component of our answer was that
some routes to reliable belief were open only to particular inquirers,
those specially equipped with background information and
44 SECT. V

inferential techniques not generally available, whereas others called

only for a grasp of an informant's language. Remembering that,
we can see how these cases, in which an inquirer has additional
information with which to offset the earlier probability, fall at least
as naturally on the 'source of information' as on the 'informant'
side of the line.

A general problem seems to arise with all definitions or analyses

of the concept of knowledge. Whatever conditions are proposed,
it seems possible to arrange for them to be fulfilled 'by accident'.
Gettier's examples, and the numerous subsequent variants on them,
do this for the analyses in terms of having good reason; examples
of deviant causal chains do it for an analysis in causal terms; examples
such as those adduced by McGinn1 do it for the counterfactual
analysis of Dretske and Nozick.
In this section I should like to investigate two questions, of which
the first is this: why is it that when we deem fulfilment of the con-
ditions to be accidental we withhold the ascription of knowledge?
It might be answered simply by saying that the concept of knowledge
requires the truth of the belief to be non-accidental. So it does,
but the answer is deficient in at least two ways. In the first place,
it makes it sound as if the point were just that the traditional con-
ditions (1) and (2) can be satisfied 'by accident', that the concept
will not tolerate this and therefore requires a third condition. Where-
as the point is more: when we add a third condition it turns out
possible to think of ways of fulfilling all three such that the existence
of the true belief, the fulfilment of (1) and (2), still strikes us as
accidental in spite of the fulfilment of (3)and when that happens
we deny that there is knowledge.
The minimal answer is deficient in a second and deeper way.
It tells us nothing about why the concept of knowledge should be
resistant to such accidental fulfilment; most concepts can perfectly
well be applicable by accident or coincidence without on that account
being any the less applicable. A bachelor, Quine has taught us,
is an unmarried man. Anything that satisfies those two conditions
is a bachelor, anything that is both a fox and female is a vixen.
Nobody is worried by the question whether, relative to the fact
of a given person's being unmarried, it is 'accidental' that he is male,
C. McGinn, this passage pp. 532-3see below, p. 58, n. 4.

or whether relative to a given animal's being female, it is accidental

that it is a fox. In cases like these there is no order of priority amongst
the conditions, no requirement that a particular one of them should
somehow connect with the fulfilment of the others and render it
Norone might think of this as a third defectdoes the minimal
answer tell us what we are to understand by 'accident' in this context.
Too much is left unspecified, too much unexplained; the hope is
that our 'practical explication' will help illuminate the fact that the
concept has this feature, whilst at the same time suggesting more
precisely just what the feature is.
What the inquirer wanted, we recall, was someone who believes
the truth about p, and has some detectable property X, possession
of which correlates well with being right about p, that can guide
the inquirer in his choice of informant. Remember also that he wants
there to be a more than merely de facto correlation between X and
the truth of the belief, for it must be legitimate for him to rely
on the correlation on this (as yet untested) occasion. Now suppose
we are satisfied, for given X and/*, that the right sort of correlation
holds; we rely on it, and are given what we later find out to have
been the right answer. But we also find out that features normally
needed to make the connection reliable were missing. As I have
said in another, related, case, what makes the connection reliable
was not, we discover, operative here. We knew things about the
informant which correlated well with being right as to whether p.
We now discover further facts and have to accept that his chances
of being right, if all that was the case, were very much lower than
we were previously lead to believe.
In what ways could the correlation between possessing X and
being right as to whether/? be unsatisfactory to an inquirer seeking
a good informant? We may start at the position in which the inquirer
believes that this correlation has held (at least nearly always) in the
past. The first problem could be that he believes that this (almost)
universal generalisation is only accidentally true, and so of a kind
that does not support an inference to the case he is at present consi-
Let us suppose that point to be out of the way, so that the correla-
tion is taken to be lawlikeit rests on something or other which
would normally permit the inquirer to extend it to the new case.
Now what else could be wrong with it? Here are some suggestions:

1. It is a correlation that you wouldn't know about unless you

yourself knew, in this instance, whether/).
2. It is a correlation that you wouldn't know about unless you
had a great deal of specialised knowledge about this particular

The first of these we can forgetunder those circumstances you

would not be inquiring. The second might be the case if in Bernard
Williams' example,2 for instance, you knew that the only way the
chairman was at all likely to come by the belief that the accountant
is depressed was via the causal route actually traversed. Another
way of putting this might be to say that the property in question,
X, would be something immensely complex, so that only someone
with minute knowledge of this particular chairman, this accountant,
and this firm's affairs, could tell that he had it. He would have
to be in a position to argue: the chairman could only come to believe
that the accountant is depressed if he is depressed himselfthe only
thing that could depress him would be the belief that the firm is
in difficultiesthe only thing that could make him think that the
firm is in difficulties would be his hearing a pessimistic report from
the accountantthe only thing that could cause the accountant to
give such a report would be his (the accountant's) depression. Then
he could argue that if the chairman believed that the accountant
was depressed there was a very high probability that the accountant
was depressed. (You might then say that there was a whole lot
more knowledge necessary, to ensure that if the chairman believed
that the accountant wasn't depressed there was a very high prob-
ability that he wasn't.) It is clear that none of these 'The only thing
t h a t . . . ' clauses would be believed except by someone with minute
and intimate knowledge of the whole situation. Furthermore,
someone with that much knowledge would virtually never believe
them, since they would virtually never all be true, even if any of
them were. In other words, this route to the conviction that the
chairman was a good informant on whether the accountant was
depressed would virtually never be open, for any practical purposes
one may confidently say never.

B. A. O. Williams, (1), p. 7.

Another possibility:
3. There is something about the present case which makes the
continuation of the correlation accidental in this instance.
Notice here, for clarity, that we must carefully distinguish between
the circumstance of the (near) universality of the correlation having
been accidental in the pastin which case no inference to the present
case would be warrantedand its being accidental in the present case
in spite of not having been accidental in the past. (This might be so
if some of the essential conditions which made it happen in the past
have been removed, but it happens nevertheless, and indeed this
would be just the structure of the standard type of Gettier-example. 3 )
We should ask, what is wrong with the correlation's being acciden-
tal in this instance, if de facto it holds? After all, the inquirer will
take it to hold, in virtue of its having held up to now, and since
it does hold in the present case (no matter how that comes about)
he is led to pick an informant who will in fact tell him the truth.
So he gets what he wanted; is there any reason to be dissatisfied?
One reply is this: that although in one sense he gets what he
wanted, that is the truth as to whether p, there is a good deal that
he would have liked to get with itand doesn't. He would have
liked, for instance, the assurance that he could use that informant
again on similar questions; but quite the contrary, he is actually
warned against it, and must now begin to look elsewhere. This reply,
however, is for two reasons not sufficient. First, it leaves one won-
dering why the fact that our inquirer did not get something else
should so cloud the fact that he did get sound information about
p. Of course, if it were clear that inquirers were never looking for
anything so specific as information, on that occasion, as to whether
p, but always for informants who were right in general on that kind
of question, then it would be clear what they felt to be missing.
But far from being clear, it doesn't even sound likely. It may be
(we shall shortly consider the question more carefully) that being
right on one issue nearly always goes along with being trustworthy
on a range of others, but why that should stop someone counting
as a good informant on the single issue in circumstances in which
that property doesn't draw the more general reliability along with
it still awaits explanation.
I mean that to be taken at face value, not as a rhetorical denial
E. Gettier.

that an explanation can be found. But whether it can be found or

not, our reply faces the more immediate charge that it relies on
a falsehood. The cases in which it turns out to have been accidental
that the chosen informant gave the right information are not by
any means all of them such as to cast doubts on his trustworthiness
on like questions on other occasions. Having been the luckless fall
guy in a Gettier case, for instance, doesn't leave one stripped of
all credit; after all, Smith proceeded according to the best canons
of evidence and reasoning. Gettier cases draw attention to the fact
that even very good reasons indeed can let you down; but if that
be allowed to tell against Smith's credentials then it tells against
everybody else's. Of course, it doesn't do either. The correlation
between having excellent reasons and being right is still what it
always was: fallible, but virtually unfailing.
Of course, there are cases of being right by accident which reveal
in their detail some misuse of the evidence, incapacity to gather
it, or insensitivity to important distinctions within it; and these do
blacken the character of the potential informant and make him for
the future in-, or at any rate less, eligible. But that all cases of being
right by accident are like this simply cannot be maintained; so it
cannot explain our response to accidental true belief in general, and
not to Gettier-type cases in particular.
However, the discovery that the correlation between possessing
the property X and being right aboutp was in this instance accidental
will have a characteristic effect on one's attitude to the very instance
in question, quite apart from what it does to one's readiness to
employ that informant on other, related, issues, or on the same
issue on other occasions. It produces that retrospective feeling of
having run a risk, of having done something that one would not
have done had one just been a little better informed at the time,
rather like finding that the person who has just driven you 50 miles
down a busy motorway without incident hasn't passed the driving
test. Believing in a disjunction (as Smith did in Gettier's second
case) on the grounds of a false belief in the first disjunct may some-
times come out right. But it can hardly be recommended as a policy.
Those writers 4 who, faced with Gettier's counter-examples, sought
to modify the 'Justified true belief analysis by adding a 'no false
lemma' clause certainly had a substantial point on their side: the
Here I am thinking in particular of Keith Lehrer and Gilbert Harman.
50 S E C T . VI

best of reasoning only preserves truth; it has no capacity of itself

to regenerate it once lost. If a train of reasoning passes through
or in any way relies on a falsehood it can only be luck if it culminates
in a true conclusion.
It may be possible to exploit these considerations about being
accidentally right to explain a prominent feature of the literature,
namely that whatever conditions (call them X) are proposed to sup-
plement (1) p and (2) S believes that p, it seems possible to find
an example in which the connection between S's satisfying X and
his being right as to whether p is accidental, and this invariably
strikes us as a counter-example to the proposal that knowledge be
defined as (l)&c(2)&tXthese conditions, whether necessary or not,
turn out to be insufficient. Now if we are looking for an X such
that possession of X gives a high probability of being right as to
p, then it is only to be expected that all analyses that specify the
X will turn out to be defeasible, for this is just what would happen
to any attempt to lay down specific necessary and sufficient con-
ditions for the practically explicated concept of the good informant.
The thought is that
S has X entails: It is highly probable that 5 is right about p,
which has the form
E entails: It is highly probable that C
But we know from consideration of the logic of probability state-
ments that unless E actually entails C itself, it is invariably possible
to find something else, D, which is both compatible with E and
such that
(E and D) does not entail: it is highly probable that C
thereby showing that E did not entail it either. This, I suggest,
is what lies behind the discovery that every analysis of 'S knows
that p' turns out to be defeasible by counterexample. The logic of
conditionals of the form
If E, then probability of C = n
is not, as the jargon goes, monotonic: only for the special case in
which nl can we add premises to the antecedent and still rely
on the consequent.
Let me try putting this another way. Offered as sufficient con-
ditions for 5 knows whether^ are:

S is right as top, and

We have seen that our practically explicated concept will have similar
structure. Of first importance will be that
(S has X) gives high probability to (S is right as to whether
For if S has X and it is not highly probable that he is right about
p then the fact of his possessing X (and merely de facto being right
about p) will not mean that he satisfies the constructed concept
these conditions will not be sufficient. An inquirer (who does not
of course know that he is right about p) should not make use of
him as an informant on the basis of his possession of X.
Now imagine someone to have proposed a specific value for X.
Note that having X must not entail that S is right as to whether
pwe have seen long ago that that makes X far too strong a condition
to be necessary. But it seems that having X must entail that it is
very probable that S is right about p, otherwise it would be logically
possible that S has X but that still it is not probable that he is right
as to whether p, and in that case he won't satisfy the constructed
concept, so the conditions proposed are after all not sufficient ones.
But it is well known that probability statements have a certain
property of def easibility which makes it impossible to find evidential
statements which, whilst not entailing a conclusion C, nevertheless
do entail that C is highly probable. It seems always possible to
find something further, D, which is compatible with , and such
that although on the basis of E alone one would regard C as highly
probable, on the basis of ( & D) one would not.
An example: when we hear that Dancing Brave has won his last
five races against top-class opposition and was clearly in the best
of health at exercise yesterday, we regard it as very probable that
he will win again this afternoon. If we add to that evidence the
further statement that Big Nig was seen in his stable this morning
tipping some white powder into his drinking-water, we cease to
regard a win this afternoon as likely. If we learn that the white
powder was only glucose we change back. If we hear that on his
way out Big Nig passed a bundle of bank notes to the prospective
jockey we lower the probability again, and so on.
What this suggests is the following: there is not going to be any

property X such that possession of it by S absolutely has to be

regarded as conferring high probability on S's being right about
p, no matter what else we may know about 5" and his circumstances;
there will always be something else which we could come to believe
(call it Y), such that (X and Y) doesn't lend much probability to
'5 is right as to whether p'. Examples of this kind will always be
available to show that the proposed analysis does not offer sufficient
conditions. The constructed concept thus turns out to have a feature
which the concept of knowledge, to judge by the recent course of
the debate, seems to have as well.
It would be premature at this stage to conclude that no set of
conditions can be sufficient for knowledge, that strictly all attempts
to state sufficient conditions must be 'Gettiensable'. Any condition,
for instance, which included the blanket proviso that there was to
be no true proposition having the properties of our D (or Y) would
seem to be immune to the above line of reasoning. We shall return
to the question in Section IX. For the moment we might explore
another path, and see if our constructed concept will resemble the
concept of knowledge in respect of what Simon Blackburn 5 has called
the Mirv/Pirv principle. He stated it as follows:
If two subjects each believe truly that p, then one cannot know, when
the other does not, unless the former is m a position with at least as much
IRV as the latter.

IRV is 'information receiver value'; it refers to the subject's capacity

to receive evidence and to think about it (if necessary) so as reach
sound beliefs. Suppose we equate this (realising however that the
equation is not completely obvious) to reaching beliefs with a high
probability of truth. Since our inquirer, in looking for a good inform-
ant, is looking for someone whose opinion on the matter in hand
has a high probability of truth, it is almost trivially true that he
will operate the Mirv/Pirv principle. For at whatever level he sets
the probability that he requires if he is to take someone as an inform-
ant, obviously it cannot arise that he will take Fred when he will
not take Mabel unless he rates Fred's probability higher than
Mabel's. If Mabel is below the line Fred cannot be above it unless
his probability of being right is higher than hers; in other words,
it cannot be right to accept Pirv when you will not accept Mirv.
S. W. Blackburn.
S E C T . VI 53

That point does not of course touch the question how we assign
these comparative probabilities in the first place, but the example
which follows in Blackburn's paper gives some indication. There
Mirv and Pirv have the same evidence, but crucial to what they
are going to make of the evidence is a certain question about the
likely behaviour of a professorand on this point Mirv has the
true belief and Pirv the false one. On the basis of this description
of the situation we give Mirv the better chance, for we quite reason-
ably take it that someone who, in reasoning, is operating with true
beliefs has a better chance of success than someone who is at some
point relying on a falsehood. What is at work here, in other words,
is the very same principle as governs the response to Gettier's exam-
ples. As with the various competing analyses, here too our approach
unifies prima facie distinct features of the literature.

The analysis of the concept of knowledge widely known as 'Reliabi-

lism' can take a number of forms. Once it has been said that know-
ledge is true belief acquired by a reliable method the question arises
what the method has to be reliable for, and a variety of answers
are in theory possible:
(a) For producing a true belief as to whether p in precisely those
(b) For producing a true belief as to whether p in circumstances
much like those obtaining;
(c) For producing a true belief as to whether p in pretty well
all circumstances likely enough to be worth considering,
(d) For producing a true belief as to whether p in all possible
All of these relate just to a particular proposition, p, and increase
in strength by stepping up the range of circumstances across which
the method is to work. We might also increase strength in another
dimension by requiring that the method work not just for the propo-
sition p but for all propositions relevantly like p. Here of course
the stringency of the requirement rises as we adjust the notion of
relevant likeness so that a larger and larger class of propositions
becomes involved.
I do not for the moment want to say anything about attempts
to decide where to stop on the scale from (a) to (d); material relevant
to that question will occur elsewhere in this essay. It is debatable,
and debated, where on the second of these dimensions is the proper
resting place for an account of knowledge; let us turn our attention
to this debate. McGinn, for instance,1 objects to Nozick's analysis
that it is concerned with reliability only with respect to the one
proposition p, and himself opts for reliability across a rather more
inclusive class. Yet there are some casessuch as that of someone
C. McGinn, p. 536.

who knows his own name but is no good at finding out anyone
else'swhich suggest that the 'single proposition' analysis may have
something to be said for it. Can the technique of practical explication
be used to throw any light on the matter?
We can arrive at much the same question from another direction,
one highly pertinent to the approach I am recommending. McGinn
makes a suggestion2 about the pragmatic value of the concept of
knowledge. Knowing that p, according to his view, involves being
good at getting true beliefs on a range of associated questions, so
'if we know that S knows that p we can infer . . . a number of truths
about the world given information about S's other (relevant) beliefs'.
Now I attempt to construct the concept by considering the situation
of someone who is looking for the truth on the sole question whether
pmy inquirer as presented is not looking for the truth on a range
of issues similar to whether/;.
This raises the question: ought I to present my inquirer as someone
who is seeking a good source of information, not just for whether
p, but for this wider class? I would have to answer the question
why he should (almost) always be doing thatthe stronger the
description of him the greater the problem of arguing that virtually
any member of any community must satisfy it. But there may be
a better route. For there seems to be at least a possibility that the
concept constructed from my original basis will include the idea
of the wider competence, or will be such as to imply this competence
in nearly all cases. Then we will be able to say that the concept
fashioned to meet the needs of my inquirer will put him in a position
to declare a bonus: not only will he have found a good informant
on the question whether phe will usually have found an informant
who is trustworthy over a range of />-like issues. Perhaps we shall
end up saying that these two turn out the same, and that looking
for ap-informant has to be equivalent to looking for a p-like inform-
ant. But that is not clear a priori, and my first guess is that these
two are not the same but only concomitant, and then only nearly
always rather than invariably.
Let us investigate the guess. What the inquirer wants, we have
seen, is someone possessing a (readily detectable) property which
correlates well with being right as to whether p. More than that,
the inquirer must believe that it correlates well, otherwise he can
Ibid., p. 540.

make no use of the fact in selecting his informant. How is he to

come by this belief?
One possibility is that he has found the potential informant to
be right on this question in the past, and so believes that he will
be right about it this time too. In this event it would be very unusual
for his belief to be limited to just this one occasion; nearly always,
if not actually always, he will believe that the potential informant
is, and will be until some drastic change alters the situation, in general
right as to whether p. But once he believes that, then he believes
something about the informant's capacities beyond the present case,
and immediately some concession has been made to the globalist
The concession isn't a very big one. So far as this argument goes,
the class of propositions over which his expertise is taken to range
might be very small indeed, though just how small depends on how
we decide to count propositions. We shall have to consider a proposi-
tion whose truth-value may change, or at least one which the inquirer
thinks may change. If we don't then our example can scarcely arise,
for the inquirer, having once determined that the potential informant
was right on that question, presumably knows the answer to it once
and for all, or at least takes himself to do so, and will not appear
again in the role of inquirer seeking an informant. But if we are
talking about something which may change, then it could be said
that what we have is in fact a changing series of questions expressible
in the same terms, amounting as it were to 'whether/? now (t\)',
'whether p now (2)', and so on. We could adopt the alternative
convention, and speak of one proposition's being assessed at different
times, in which case it would sound as if our reliabihsm was of
the local rather than the global brand. Nothing really turns on this
choice, of course; the solid point is that the inquirer's confidence
in the informant will always in some measure extend beyond the
single case which is his current concern.
The extension is too minimal, however, to give much support
to McGinn's objection to Nozick.3 Stronger support may be on
the way, but before we look at it we may note that there are at
any rate some cases in which there could be knowledge that/) when
the relevant class of 'p-like' , truths doesn't reach beyond p itself,
assessed at other times. Significantly, this occurs when the informant

Ibid., p. 536.
S E C T . VII 57

stands in a unique relationship to the fact in question, or may be

presumed to be uniquely familiar with it. An infant may know what
its name is without knowing what anyone else's name is, and without
knowing its age, and it may know its forename without knowing
its surname. Being confident that someone knows their name doesn't
help us to judge that they are valuable informants on many other
questions, if any.
If we try hard enough we will probably be able to imagine circum-
stances in which we might reasonably trust a potential informant
on the question whether there was a table in the next room, but
not on the question whether there were chairs there, or anything
else about the furniture. (In fact one doesn't even have to try very
hard: suppose that he has never been in the room, but has just
seen a table being carried through the door.) So if we are looking
for strictly necessary conditions of being a good informant on any
question whatever, we shall plump for localism and confine ourselves
to the individual proposition at issue. But it is clear that all these
examples are in some way exceptional, either in point of the proposi-
tion in question and the informant's particular relation to it, or of
the circumstances in which he acquires the information. Far more
common will be the case in which we trust him over p because
we believe (or have found) him to be good at discerning the truth
over a range of related matters. From up there in his tree Fred is
good at telling where things areindeed, people in general are good
at itit isn't any particular aptitude for spotting sabre-toothed tigers
that we are relying on. If you are well placed to see enemies you
are also well placed to see prey. In the vast majority of cases, if
Fred is good for information on p, he will be good for information
over a range of similar propositions. What counts for these purposes
as similar to p will vary widely, but it is something which we are
mostly pretty good at judging. If what makes me think that Fred
is good for information on the colour of the mat is that I can see
that he is looking at it, I will take him also to be good on the
question whether the cat is on it or not; if he is in the next room
and is a good informant about its colour because it is his house
and his mat, I won't expect his reliability to extend to that question,
but rather to others of a different type.
One could put it like this: on the basis of our estimate of the
potential informant's capacities, situation and circumstances, we
form an overall estimate of his belief-acquisition potential with

respect to p. In nearly all cases the information we have used to

estimate his potential with respect to p will justify an estimate of
his potential with respect to quite a lot of other propositions as
well. On different occasions (witness the above example) the factors
that go into estimating his potential with respect to p may be differ-
ent, although p is the samethey may be perceptual capacities on
one occasion, inferential on another, a matter of his memory on
a third, and in each case different considerations about his situation
and circumstances will be pertinent. One effect is, of course, that
the range of other propositions associated with reliability as to
whether/) will change accordingly; there will not be a static chunk
of further information which invariably goes along with knowing
whether p.
That thought necessitates an adjustment to what McGinn says
about the pragmatic value of the concept of knowledge. It is not
that once we know that Fred knows that/? we can start using Fred's
beliefs as to q and r as reliable indicators as wellthe premise that
he knows that p is too weak for this purpose. We have to know
also the circumstances under which he came by his true belief as
to whether p, since only this additional knowledge will allow us
to judge which other propositions are so to speak 'adjacent' to p
in this case. If you take me to know that there is a piano next
door because I know the house, then other propositions about the
furniture of that room are adjacent; if I know it because I heard
the piano through a closed door, then those other propositions are
not adjacent at all. It isn't, therefore, the concept of knowledge
which has this pragmatic value; it is what one knows, on particular
occasions, about how the knowledge was acquired.
This does something to explain a phenomenon that McGinn has
latched onto:4 the fact that one can generate counter-examples to
Nozick's analysis by inventing cases in which the reliability (or 'track-
Ibid. p. 536, with reference back to p. 533. A quotation may help the reader:
'Suppose you are surrounded by straight sticks immersed in water that therefore
look bent to you; you, however, take them to be in air, and so you falsely believe
of each of them that it is (really) bent. There is, though, one stick that is not immersed
in water and it really is bent; on the basis of how that exceptional stick looks you
believe it to be bent. Again, I take it that you do not know that that stick is bent,
since, in view of the circumstances, your belief is only accidentally true. But Nozick's
variation condition is satisfied in this case, since if that stick were not bent you
would not believe it to be, because it would not, being in air and not water, look
bent. So here we have a perceptual belief that tracks the truth of the believed proposi-
tion but does not rank as knowledge.'

ing') holds for p in particular but not for other propositions 'very
like' p. Nozick's conditions are then satisfied, but we are unwilling
to attribute knowledge of p. The example must be so constructed
that the putative knower's capacities, and those features of his situa-
tion which might be thought to be making him reliable as to p,
are such that one would be led to expect reliability with respect
to these others as well. When it turns out that the subject is not
reliable with respect to them, this makes us feel that it cannot really
be those capacities etc. which are operative in respect of p, in spite
of the fact that his belief 'tracks' its truth-value. We needn't therefore
say that the reason why this is not knowledge is because knowledge
of p calls for knowledge of the other ('jp-like') propositions as well;
we can say that even with respect to p the subject is not really
exercising those capacities which make him reliable, so that we have
to regard it as a fluke that he meets Nozick's conditions. And the
trouble with flukes, from the point of view of the inquirer, is that
they aren't predictable; which makes them of no use to him, since
he needs to be able to tell in advance that there is a high chance
of learning the truth.
This is paralleled by the line of thought that arises if we go back
and look at McGinn's case from the perspective of our inquirer:
if we now consider someone trying to use that person as an informant
about whether p (whether a certain stick is bent or straight), we
can see that they will judge him to be a good informant because
they will see him looking at the objects in question, and since the
properties he is judging them to have are (we suppose) simply percep-
tually detectable they take it that he possesses a property correlating
well with being right as to whether p. What they learn, however,
when they learn of his incapacity to give the right answer to the
other questions (whether the other sticks are bent or straight) is
that there is special reason to mistrust the correlation in this particu-
lar case. The question whether theywho don't at this stage, let
us remember, know whether/" or notshould regard him as a good
informant is back in the melting pot. Unless they can find some
further property which he possesses and which (i) correlates well
with true belief as to p and (ii) doesn't correlate equally well with
true belief as to all these other propositions about whichas we
have now found outhe doesn't hold true beliefs, then they must
suppose that his chances of telling them the truth about p are, for
all they can tell, no better than random. And then, since their own

chances are presumably at least that good, they must cease to regard
him as a potentially useful informant.

It has become common to distinguish between 'internalist' and

'externalist' accounts or analyses of knowledge. If we think of such
an account as laying down necessary and sufficient conditions for
knowing thatp, first the truth of p, second the belief that/?, third
some further X which for these purposes we need not specify more
closely, then an internalist will maintain that the knowing subject
must have some kind of awareness of the fulfilment of the third
condition, whereas an externalist takes it to be enough that the third
condition hold, even if no thought of that condition, let alone belief
about it, has ever crossed the subject's mind.
There is a widespread habit of thinking of some analyses as intrinsi-
cally internalist, others as intrinsically externalist. That, at any rate,
is what I suspect, so I shall allow myself a paragraph to point out
that it is not quite the case. The analysis in terms of justified true
belief, or having evidence, or good reason, is internalist (so the story
might go): having good reason to believep involves having an aware-
ness that some proposition q, also believed, bears a confirmatory
relationship to p. On the other hand, the causal analysis (to take
one example) only requires that the causal connection between the
circumstance that p and the belief that p should in fact obtain. If
the subject knows, or believes, that it obtains, well and goodbut
that is superfluous to the question of his knowledge that p. And
the position is the same with the tracking analysis, reliabilism, and
Unger's non-accidentalism. l
But in fact there is nothing intrinsic to any of these analyses which
demands an internalist or an externalist treatment, and if they are
attached to internalism and externalism in the way I have indicated
that is more a matter of history than of their inner logic. Thus
whether the classical 'true belief with a good reason' is internalist
or not depends upon how one goes on to treat the notion of having
good reason. Many philosophers, clearly, have taken it to involve
P. Unger, (1).
62 SECT, V I I I

awareness, on the part of the subject, both of the reason and of

the fact that it is a good reason; but if one follows Dretske,2 for
instance, in analysing the possession of good reason in terms of
the fulfilment of Nozick-like counterfactuals, then one finishes with
the tracking analysis of knowing, and externahsm at once becomes
an obvious option. I say an option, rather than obligatory, because
although the tracking analysis is standardly formulated externalisti-
cally, nothing about it prevents us, if we are so minded, from adding
a clause to the effect that the subject must be aware that the counter-
factuals hold; it is just that, recently, people haven't been so minded.
Of course, there is one very serious problem for internalism: if it
allows the awareness of the fulfilment of the third condition to
become too strong, in other words to begin to look like a demand
for knowledge, it is threatened with regress. But that is not a problem
that arises for some particular analyses and not for others; we meet
it once we try to take any analysis in an internalist version. And
this we can do: it is worth remembering that Goldman's original
statement of the causal theory spoke of the need for the subject
to be able to 'reconstruct' the causal chain in its important links
which introduces an internalist element, at least on the obvious inter-
pretation of what such reconstruction would amount to.
At first sight it looks as if our approach is bound to favour externa-
list theories: will not the characterisation of the good informant
be externalist? After all, if we judge Fred to have a property that
correlates well with having the truth as to whether/?, we shall judge
him to be a good informant quite independently of whether he is
himself aware of having that property or of its correlation with the
truth of his belief.
Before accepting that, there is one possible counter to it that we
should dispose of. Earlier, in Section II, we wondered whether a
good informant might not need to be someone who offers his opinion
confidently. Should the question be raised anew at this point in
the argument? Might not that confidence call for awareness that
one possesses a property correlating closely with being right on
the matter in hand? If so, would that not put internalism in an
analysis much on a par with what was then under discussion: the
inclusion of the belief condition?
If we are really thinking of an internalist analysis, in which the
F. Dretske.

subject's awareness is to be laid down as a logically necessary con-

dition, the answer would surely be negative. Confident belief can
certainly exist in the absence of awareness of those properties of
oneself which justify the confidence, or make it very likely that
the belief is true. Things may not be quite so clear cut if we take
the question strictly as put, that is as asking whether internalism
may not have the same kind of status as the belief condition; for
doubtless in the vast majority of cases people are aware of some
such property, and this (contingent) fact about confident informants
must have an effect on the way in which we think of them. (This,
according to the argument of Section IV, is what underlies the plausi-
bility of the traditional analysis in terms of justification or reasons.)
But be that as it may, we have still found one point on which the
supporter of the belief-condition can call which has no analogue
for the internalist: the lack of belief seriously disrupts our willingness
to think of the subject as an informant rather than an evidential
source of information. In part, at least, this happens because we
are then reluctant to regard him as properly telling us whatever
it is that he says. The failure of the internalist condition, on the
other hand, has no tendency in that direction.
So the view that our approach will at least in some degree favour
externalism is right, I think. But perhaps it is only right so long
as we stay with the third-person perspective, and concentrate on
the case in which someone is trying to decide of someone else
whether or not they are a trustworthy informant. In some respects
at least the position changes if we look at the situation that obtains
when someone has to make this decision about themselves.
At the beginning of this essay I argued that consideration of the
third-person case was the sounder starting point. If one finds oneself
with a belief as to whether p, I said, that pre-empts inquiry; no
search for an informant will begin. But I also hinted that slightly
more sophisticated states of mind might bring the first-person ques-
tion into focus; and it can easily be seen how, in the primitive situa-
tion that we first imagined, such states of mind could arise. For
one thing, there will be circumstances in which a group is looking
for an informant on some issue: who knows whether p? And it
will be important to have a practice whereby people sometimes
declare themselves to be qualified, since it will often arise that they
are themselves the only person in a position to tell whether they
are qualified or not. That would be so, for instance, if the question
64 SECT. V I I I

relates to what happened at a certain place at a certain time; it may

well be that the only person aware that Fred -was there then is Fred
himself. Or again, it may well be that nobody but Fred is aware
that he has just seen a signpost.
Now an immediate reaction might be that all Fred needs to do
is to offer an opinion as to whether p, either to assert that p, or
to assert that not-jf, and that this doesn't involve turning the concept
of knowledge, or of being a good informant for the group to follow,
on himself. Sometimes, perhaps, but there will be many cases in
which the bald assertion only provokes further challengeshow
do you know, what makes you think that, Fred?and the fact that
this is likely will induce people to consider their defence in advance,
which means considering whether or not they themselves are inform-
ants acceptable to the group. Then it is perfectly understandable
that they should often announce themselves as such in the first place.
Who knows whether/?? And Fred replies: I do. And there is another
reason why the practice of asserting (or denying) p cannot always
do the job: the group may wish its attention drawn to a sound
informant on some general topic, before any particular question
has been decided upon.
In addition to that, we are not infallible. Some of our beliefs are
false, and what is more, we pretty soon make that discovery for
ourselves, even if others don't force it upon us first, which they
very well may. So we shall develop a practically based interest in
whether we ourselves are good informants; as in the third-person
case, the interest will be sharper where the practical importance
of having a true belief is greater. We shall sometimes be interested,
therefore, in asking of beliefs that we have, whether they are to
be relied upon; and, perhaps more often, in cases in which we want
a belief as to whether p but don't have one, we shall be keen to
acquire one in a way which makes it likely to be true when we
get it. And in this latter case it may be much more convenient to
set about making ourselves good informants directly than to try
to find someone else who is one. One great advantage of using
yourself as an informant, of course, is that unlike everybody else
you are always around; and in any case you are sure to be the
first person you ask.
There is therefore nothing about the method of practical explica-
tion which disallows consideration of the first-person case; on the
contrary, it can be seen to be a legitimate, indeed a required, exten-
SECT. V I I I 65

sion of the approach. But once we do consider it we can see immedi-

ately how the tendency to internalism arises. Internalism consists
in adding to the third condition (or so interpreting it that it implies)
a clause to the effect that the subject is aware of possessing Xhaving
good reason for his belief, or whatever. The self-directed, or first-
personal, inquiry brings us into the neighbourhood of internalism
because it forces on us the question 'Do I meet the third condition?';
and to decide that we do, or that we are good informants, we must
satisfy ourselves that the answer is 'Yes'. Then we are in just that
extra state which internalism characteristically adds to the externalist
I say 'brings us into the neighbourhood of rather than 'brings
us to' internalism because the fact that we are in that extra state
doesn't oblige us to build it into the concept. But the self-directed
nature of the investigationit is ourselves we are aiming at certify-
ingbrings the extra state on to the scene in a particularly seductive
way: I cannot get into a position to certify myself a good informant
with respect to p unless I consider whether I meet the condition
X and decide positively. And this is easily confused with the thought
that being a good informant involves being aware, or at the very
least believing, that one fulfils the third condition. In the third-
person investigation the opportunity for the confusion does not arise.
If I am to certify Fred a good informant then I must satisfy myself
that Fred fulfils (3); but to confuse that with the claim that Fred
must satisfy himself of that as well is (happily) too gross to be trouble-
In case any reader is unconvinced that the former confusion is
tempting, I shall say a few words about a parallel confusion which
can easily be documented. Not only is it parallel, it is closely
relatedindeed it may at bottom even be identical. I refer to the
much discussed question of epistemic logic, whether or not knowing
that p entails knowing that one knows that p. I shall call it the
'Iteration Principle': Kp entails KKp. It is not difficult to see that
this question is very closely related to the one we have just started
from, namely whether to give an internalist or externalist analysis
of knowledge. If we think schematically of an analysis of knowledge
as consisting of the truth condition, the belief condition, and some
further (perhaps complex) condition which we refer to as '(3)', then
Internalism is characterised by the thesis that to know that p the
subject must know that (3). The Iteration Principle holds that too,
66 SECT. V I I I

but adds that there must be knowledge of the two preceding con-
ditions as well.
That an externalist will return an immediate' 'No' to Iteration
is clear enough; just a little more thought is needed to appreciate
the force of the temptation for the internalist to answer 'Yes'. If
to know that/? is to satisfy the conditions:
(2) S believes that/;
(3) S has X
then knowing that one knows is to know that the three conditions
hold, that is to say:
(!') S knows that/;
(2') S knows that S believes that/?
(3') S knows that S has X
Now if an internalist so understands (3) that he does not really
distinguish it from (3'), then little else is needed to effect the passage
from knowing, { 1 & 2 & 3 } , to knowing that one knows: {!' &
2' & 3'}. Since (3) is true, our internalist will take it that (3') is
true. Since ex hypothesii the conjunction {1 & 2 & 3} holds, and
he takes that conjunction to define knowing, he will take it that
(!') holds. Now only (2') remains to be accounted for. Our interna-
list might take that step as a result of believing something like the
traditional thesis of the transparency of our mental states to con-
sciousnessthat we always know what we believe. But without
accepting that dubious thesis, he might get to (2') as a common-sense
consequence of believing (3'). For if his chosen A'is 'has good reason
to believe that />', and S knows that he has itwhich is what (3')
saysthen S is hardly likely to miss the fact(2)that he believes
that/;. So (2') will hold, and with it the full set of conditions for
knowing that one knows. If the chosen X,, as in the causal theory
or the tracking analysis, actually mentions the belief that/7, the pas-
sage to (2') is even more obvious. So it should not be in the slightest
bit surprising if the internalist takes the view that 'S knows that
/>' and 'S knows that S knows that p' are equivalent (though he
might, because of the status of (2'), resist the conclusion that the
equivalence is strictly speaking a two-way entailment).
Now at least with regard to the 'If Kp then KKp' principle we
can certainly find examples of just the confusion under discussion.
SECT. V I I I 67

Here are two from Hintikka's Knowledge and Belief, one quoted
(approvingly) by him, the other actually used. 3 Schopenhauer, in
On the Fourfold Root.. ., writes:
. .. just try. . . to know without knowing that you know .. , 4

a challenge the force of which must surely rely on the hope that
your attempts to know will include attempts to assure yourself that
you know, and the further hope that this will convince you that
knowing itself includes knowing that you know. The joker here,
of course, is the second word. If we try to know that p we are
going to concentrate our minds on the question whether we know
that p and what we have to do to bring that state of affairs about.
Then, if our efforts go well, we very likely shall finish up knowing
that we know that p. At least it will be the case that we know
that p, we shall believe that we know that p, and we shall be in
a state such that people in that state who believe that they know
that/) virtually always believe truly.
But it is not the knowing that has produced the iteration; it is
the trying. Think of someone who, instead of trying to know that
p, tries to find out simply whether p. If he does the job well, he
will end up knowing that p (or not-/? as the case may be). At any
rate, he will arrive at a true belief and a state which correlates well
with truth in beliefs of that type. But whether he will know that
he knows is quite another matter, depending on how self-conscious
his investigation of/> was; he may, but quite likely he won't.
Hintikka himself uses a similar route:
. . . all those circumstances which would justify one in saying 'I know'
will also justify one in saying 'I know that I know'. 5

Now the joker is justification. Does it smuggle in internalism? Am

I justified in saying T know that p' just because I do know that
/>? Or, as is plausible, is something else needed? For if it is, it may
be the something else, and not just the fact of my knowledge that
p, that entails the knowledge of the knowledgeassuming, that is,
that anything about the situation has this consequence, which is
far from clear. What is, I think, clear, is that if I set out to be
justified in saying 'I know that p' I will almost certainly aim at
J. Hintikka. See pp. 108, 111.
A. Schopenhauer, ch. VII, Sect. 41, cited in J. Hintikka, p. 108.
J. Hintikka, p. 111.
68 SECT. V I I I

getting into a position in which even the internalist would agree

that I satisfy the conditions for knowledge; and if successful I shall
end up knowing that I know. But that has nothing to do either
with the question whether mternahsm is correct or with the Iteration
Principle. It is after all true for any value of p, not just for ones
that contain an occurrence of 'knows that', that if I set out to be
justified in saying that p I shall (assuming competence) very likely
finish up knowing that p (unless, question-beggingly, internalism
be assumed); but it does not follow that p can only be true if I
know that/?. Omniscience does not come that easily.
We have seen that there will often be motives, arising in the primi-
tive situation or state of nature, to apply the concept of the informant
to oneself. This does not mean that there is anything fundamentally
first-personal or fundamentally third-personal about the concept of
knowledge. The informant whose credentials are under scrutiny may
be either oneself or another. But this very impartiality between first
and third person perspectives does, I think, speak in support of
those who offer externalist accounts of the concept. The third-person
view certainly favours externalism, whereas the hint of internalism
experienced in the first-person case gives no sufficient reason to
break the semantic symmetry and posit a different, because interna-
list, concept that one applies to oneself. For the hint of internalism
can be accounted for without supposing that one is then asking,
of oneself, a question with an internalist content. Simplicity and
elegance favour reading the questionam I a good informant with
respect to p?externalistically, and seeing the internalist aura as
an unsurprising consequence of the fact that it is here being asked
internally, that is to say, of oneself.

The concept of the good informant, I have argued, has the property
that once we try to state specific necessary and sufficient conditions
for it, once we add some specific condition Y to the requirement
that the informant believe the truth on the matter at issue, it proves
possible to 'outflank' the attempted analysis: we can imagine further
facts, compatible with the informant's possessing F, such that when
they are included in the picture his rating drops to an unacceptable
level; so the suggested conditionstrue belief plus Ywere not,
after all, sufficient.
We noted that the recent literature on the analysis of the concept
of knowledge suggests something suspiciously similar: when a speci-
fic analysis is proposed, sooner or later someone devises a counter-
example to demonstrate the insufficiency of the proposed conditions.
A very general and approximate description of these counter-exam-
ples might be: they all present possible circumstances in which the
further condition X is satisfied, but satisfied in such a way that
it becomes accidental that the subject holds a true belief as to whether
p. What, with more precision, does this really amount to?
Gettier addressed himself to the 'justified true belief (JTB) analy-
sis. A condition of the construction of his type of counter-example,
he pointed out, was that it should be logically possible to have
a justified belief in a proposition that is in fact falseS is justified
in believing that p is not to entail thatp is true. This being granted
(it is not easy to resist it without having to admit that justification
is an unnecessarily strong requirement) Gettier proceeds to invent
situations in which the subject is indeed justified, but in which,
given the manner in which the justification arises, the likelihood
that he is right as to whether p is not thereby increased.
This makes the JTB analysisgranted Gettier's principle about
justificationstructurally very similar to the conditions on a good
informant: true belief plus possession of some property which lends
high probability to truth of belief, but does not (because such a
70 S E C T . IX

requirement would be too strong to be a necessary condition) entail

truth of belief.
It seems plausible to say that a third condition which did not
lend high probability to truth of the belief could not (taken with
the earlier conditions) be sufficient for knowledge. It looks, in other
words, as if a crucial part of the role of the third condition (whatever
else it may need to do, if anything) is to provide this probability.
That being granted, it can be seen that such an analysis will always
be vulnerable to counter-examples that work by adding further cir-
cumstances compatible with the third condition and which, when
added, leave that probability unacceptably low. To this we must
add the caveat, however: provided the third condition in question
does not entail that the subject's belief is truefor then no such
further circumstances will be conceivable.
The caveat suggests two continuations. Not only the JTB analysis,
but the other three principal contenders as well seem to be indefi-
nitely open to 'non-sufficiency proofs' by counter-example; so we
shall have to ask whether they share with it the relevant property
that their proposed third conditions do not entail the truth of the
subject's belief. And as a preliminary it might be as well to ask
whether the JTB analysis may not have an acceptable version which
lacks that property, that is, one which uses a notion of justification
according to which justification entails truth.
The first thought, already mentioned, is that any such concept
of justification -would be far too strong to be necessary. Using it
in the analysis would amount to the demand that the subject be
in possession of evidence which entails the truth of p. But that
demand is met in so few cases that the analysis would decisively
rule out virtually everything, and such a misfit between what actually
is knowledge and what is normally taken to be knowledge is drastic
enough to kill the proposed analysis.
For the time being, I think, we have to agree with that. Admit-
tedly, until we have decided what features of use and usage are
constitutive of a concept it would be premature to decree it unthink-
able that nearly every application of some concept should be wrong.
But someone who wishes to analyse knowledge in such a way that,
if he were right, that would actually be the case, surely owes us
some half-way plausible hypothesis about how the massive misfit
could have come about. If it were 'witch' that we were talking about,
then the debt could be settled: there is a very naturalthough some-

what pathologicalexplanation of the fact that the word acquired

such a meaning as not to apply to any of the many persons to whom
it was applied, and was applied to them nevertheless. But the concept
of knowledge doesn't look anything like so tractable, and until there
is a story about it having something approaching the plausibility
of the obvious one about 'witch' we should provisionally assume
the worst.
It might be suggested, however, that the concept of justification
be handled rather differently. Could we not say that a lower standard
of evidence than deductive entailment of the conclusion is sufficient,
but that it is also a necessary condition of justification that the belief
actually be true? So even very good evidence for/? would not pro-
perly be called justification of the belief that p if that belief turned
out to be false. If this were the concept of justification used as the
third condition of the JTB analysis then the third condition would
entail truth of belief, but knowledge "would not call for evidence
so good as to entail the proposition known.
There is no escape down this route. It makes the concept of justifi-
cation as it were bipartite: (i) S has good evidence for p, (ii) p is
true. So the JTB analysis, using this conception of justification, ren-
ders 'S knows thatp' as:
(1) p is true
(2) S believes that/?
(3) (i) S has good evidence for/?
(n)/? is true
Since (3)(ii) duplicates (1) its effect is merely nominal. The same
counter-examples as Gettier used against the sufficiency of JTB as
he wished it understood can be used again, with exactly the same
result. But there is another modification to the JTB analysis, widely
known as the 'no false lemma' principle, which is far from nominal.
Since the salvage which it promisesif it works at allextends to
other types of analysis than JTB, I postpone consideration of it
until the end of the section. For the moment we may move on
to look at the recent alternative analyses, the causal theory, the
tracking theory and reliabilism.
To the true belief that/? the causal theory adds a causal connection
between the belief and the fact that p. It mayin fact mustgo
on to place restricting conditions on the sort of causal connection
allowed (it must not be 'deviant'), or on the subject's position in
72 S E C T . IX

regard to it (Goldman: he must be able to 'reconstruct' its essential

links correctly). But whatever these restrictions are they do not
affect the point that the existence of any causal connection with
the fact that/> entails thatp is a fact, hence that the subject's belief
that p is true. So the third condition does not merely make the
truth of the belief probable; it entails it. The structural parallel with
the JTB analysis is lost, and with it our neat explanation of why
the causal analysis too seems to be indefinitely counter-exemplifi-
The same is true of the Nozick-Dretske analysis, which adds
the two counterfactual conditionals to the requirement of true belief.
They entail that the subject's belief that p is true, not merely that
its truth is highly probable. Yet again we find it possible to invent
odd (or 'deviant') ways of satisfying them which indicate that they
do not combine with true belief to give logically sufficient conditions
for knowledge.
Reliabilism adds to true belief the need for the belief to have been
formed 'by a reliable method'. Here various well-known questions
arise. First, it is not clear how we determine the method by which
a given belief was formed, for there will have been a highly complex
process which can correctly be described in many ways. Under some
descriptions it will be reliableacquisition of the belief that p by
that kind of process will (nearly) always lead to truthunder others,
equally correct descriptions of it, it will not be. The easy way out,
to say that a belief was acquired by a reliable method if there is
some true description of the process under which it is reliable, is
not as easy as it looks, for there will always be some description
of that kind unless we make further provisions setting limits on
what is to count for these purposes as a description of the process.
Thus 'guessing' might be a description of the process, but 'guessing
correctly' had better not be, even when the subject did guess cor-
rectly. (Even the way in which Bernard Williams's company chair-
man1 came to believe that the accountant was depressed was, if
described in sufficient and properly chosen detail, a way of coming
to that belief which would always lead to truth.)
Suppose we have solved this problem, so that we know -which
descriptions of a process of belief-acquisition count as specifying
'the method(s)' by which the belief was acquired. Then the process
See Section VI.

will be as reliable as its most reliable specification, that is the one

(or 'ones' if there is a tie) under which it has the best chance of
issuing in a true belief. The next question is: how good does that
chance have to be for the process to be 'reliable'? Must every process
satisfying that description lead to a true belief, and if so are we
speaking of every actual, or of every possible process ? Must it (either
way) be every process, or will it do if it leads to a true belief in
the vast majority of cases? Are we only speaking of the cases in
which it is applied to settling the precise question whether p, or
are we also thinking of investigations of other propositions of the
same type asp? If the latter, how do we decide which propositions
are 'of the same type' asp?
Fortunately we don't, for the moment at least, have to decide
all or any of these questions on the reliabilist's behalf. Unless he
calls for a method so reliable that it would lead to a true belief
about p on any possible occasion on which it might be applied to
that question then it will always be possible to think of further
circumstances such that, even though the subject had reached his
belief about p by a reliable method, his chances of being right about
p were not (given these further circumstances) thereby improved.
And this will leave reliabilism open to counter-examples showing
it not to offer sufficient conditions for knowing.
Are there any prospects for the reliabilist in taking the stronger
line and calling for a method that would yield the truth about p
under all circumstances, possible as well as actual? In theory it would
be open to him to try something analogous to our JTB-theorist
who wanted to make justification a matter of two logically indepen-
dent components, one being possession of adequate evidence and
the other the de facto truth of the belief. What this comes to is
that identification of the method would include the fact that it had
succeeded; so that any attempt which ended in a false belief, however
similar to the first in other respects, would be said not to involve
the use of the same method.
Quite apart from the weirdness of such a use of 'same method',
this clumsy manoeuvre achieves nothing, for the reason that we saw
in the parallel treatment of 'justification'. The reliabilist has now
defined knowledge like this:

(1) p is true
(2) S believes that/)

(3) (i) S acquired the belief thatp by doing X

(ii) S's belief that p is true
where doing X is something that correlates well with, but does not
entail, forming a true belief as to whether/;. And once again (3)(ii)
is redundant, since it just recapitulates the conjunction of (1) and
(2). The same counter-examples as would demonstrate the insuffi-
ciency of {(1) & (2) & (3)(i)} do the same after the addition of
(3)(ii)for (3)(ii) adds nothing.
There are, however, some descriptions of processes leading to
belief that entail the truth of the belief without falling apart into
two components, one not entailing the truth of the belief and the
other just explicitly stating that the belief is true. Many beliefs, we
all suppose, have as an essential part of their cause the state of affairs
which is their object. As we have seen, this description of the process
of belief-acquisition does entail the truth of the belief acquired; and
it does not decompose into two independent conditions, such as:
(3) (i) S's belief that/) was caused
(ii)p is a fact
So it seems that a reliabilist might say that a 'reliable method' was
indeed one that couldn't fail, and allow as productive of knowledge
only such processes as are susceptible of that kind of description.
He would admit causal processes involving connection -with the fact
believed, processes that 'would lead to the belief that/; if and only
if/)', processes involving the due consideration of reasons logically
sufficient for p, and perhaps not much else, unless he could think
of some new and plausible analysis. What he is doing, clearly, is
borrowing (in one case in strengthened form) the analyses already
in circulation and disjoining them. To summarise these possibilities:
(1) Reliabilism calls on processes which have (only) a very good
chance of delivering true belief.
(2) Reliabilism disjoins the existing analyses, in one case (JTB)
taking a strengthened form of the third conditionand thus
achieves a logical guarantee of truth.
(3) Reliabilism suggests that there are other descriptions of pro-
cesses leading to belief such that any process satisfying them
must (logically) deliver truth.
In the case of (1), as soon as anything specific is suggested there

will appear 'Gettier-like' counter-examples, for the reasons given.

In the case of (2) there is nothing new to look at. In the case of
(3) we must wait to see what these descriptions are, bearing in mind
the while that (a) there will be no point to them if they are of the
sort that build in the truth of the belief as an independent condition,
and (b) that the reliabilist must not overstretch what counts as the
description of a processon pain of having all true belief turn into
That out of the way, at least until further notice, we can return
to the causal analysis. Here the third condition (C) does entail (not
just give high probability) that S's belief that p is true. So when
the conditions turn out insufficient that cannot be explained just
as the concept of the good informant suggests, by saying that in
the counter-examples we are adding further compatible information
(FCI) such that {(FCI) & (C)} does not lend high probability to
'S is right that p''. Since in this case (C) entails 'S is right that p',
so does the conjunction of (C) with any (FCI).
Nevertheless, I believe that the original intuition is correct: what
bothers us about the 'deviant causal chains' examples is that they
present cases in which the attainment of a true belief that p, even
by a process causally connected with the fact that p, is a fluke or
accident. In some sense, it was very unlikely to turn out that way.
We may look at another counter-example proposed by Bernard Wil-
liams in the paper already cited. Admittedly, Williams characterised
it precisely as a case in which the truth of the belief was not an
accident, but this shouldn't put us off; it just shows that the concept
of being an accident is a slippery oneI shall argue for a different
application of it. First the example:
A, being from Guinea, tells B falsely that he is from Ghana; but (let us
fancifully suppose) owing to features of A's spoken English which are pecu-
liar to Guineans, B takes him to have said 'Guinea' when he said 'Ghana'.

Now B has come to believe (what is true) that A is from Guinea,

and (so Williams continues) 'it is no accident, relative to A's being
from Guinea, that this has come about'. What we can certainly agree
is that the fact of A's being from Guinea plays a causal role in the
process leading to B's beliefthat was how A came by the peculiarity
of accent. But can we not also agree, looking at the process, that
there was something deeply accidental, or coincidental, or unlikely,
about B's having arrived at a true belief rather than a false one?

First, A had to choose the spoken word for conveying his infor-
mation; secondly, when he shaped up to tell his lie he had to hit
on precisely that country the English name of which would sound
to B like 'Guinea' when pronounced with a Gumean accent. Nigeria
wouldn't have done, nor would Ethiopia or Zimbabwe or (probably)
anything else, butit just happened to be 'Ghana' that he picked
on. Well, well! Relative to the evidence in our possessionand that
is what judgements of probability are relative toB was extremely
unlikely to acquire a true belief. But the unlikely can happen; and
this time it did.
Make the comparison: let there be a chair here, and suppose that
I look at it attentively. Now, what has to happen for me to fail
to acquire the belief that there is a chair here? Answer: something
pretty unusual. Let there be a Guinean here intent on deceiving
me about his origins, and suppose that I listen to what he tells
me and believe what I understand him to say. What has to happen
for me to acquire the right belief (without benefit of knowing that
he is lying, without benefit of experience of the Guinean accent)?
Answer: something pretty extraordinary. Wasn't he unlucky? Of
course he wasthe odds were piled house-high in his favour.
We find much the same if we look again at one of Williams' other
examples, that of the Chairman of the Board. People who are
depressed characteristically have a certain type of facial expression,
a certain heaviness of bodily movement, a certain style of speech,
an absence of energy and initiative, a certain 'slant' on life and its
questions. Someone who detects these features in the normal way
and infers, if he finds them, that their possessor is depressed, has
a reasonably good chance of forming a true belief. But our chairman
didn't do this; he came by his (true) belief only because the account-
ant, of all the ways in which he might have manifested his depression,
happened to come up with one (of the very few, we may presume),
which had the effect of making him (the chairman) depressed. And
he had to come up with it for the reason that he (the accountant)
was depressed, rather than for the reason which must be at least
equally common: that (in his everyday professional state of mind)
he had seen that the firm's state was actually parlous. The chairman's
chances of ending up with a true belief would have to be assessed
as very slender by anyone who did not already know the whole
story, including the facts about the accountant's state of mind.
What seemed to cause a problem here was the wrong approach

to chance. Of course, given the premise that I acquired the belief

that p by causal connection with the fact that p we can conclude
deductively that I acquired a true belief. But whether it was only
to be expected that I would acquire a true belief, or whether on
the contrary I needed the luck of the devil himself, is a question
not touched by that premise or the deduction we can trivially make
from it. What the counter-examples do is move the believer to the
lucky end of the scale; they present situations in which he gets things
right against all the odds. 'Deviant' causes produce counter-examples
to the sufficiency of the analysis precisely for that reason, and their
permanent availability is due to the very feature of probability judge-
ments which, as we have seen, has just the same effect on any attempt
to specify conditions logically sufficient for qualifying as a good
It is now time to redeem a promise given earlier m the section.
In Gettier's original counter-examples to the sufficiency of the JTB
analysis, a subject reasons admirably but reaches a conclusion which
is in fact false. From this falsehood he then draws (with unexception-
able logic) a further inference which happens to yield a truth. The
ensuing discussion of such cases therefore lead very naturally to
modification of the analysis by addition of the 'No false lemma'
(NFL) principle associated in particular with the names of Keith
Lehrer and Gilbert Harman. 2 Not only must the subject have good
reasons for his belief; the train of reasoning which leads him to
it must not include anything which is, in fact, false. Thus we cut
out Gettier-cases at what looks very like the root.
It soon becomes clear that the principle itself had better be modi-
fied a little. As stated above it can only be applied to cases in which
we are able to identify the particular train of thought through which
the subject passed in getting to the belief. So a reformulation in
conditional terms comes to be preferred: there must be no proposi-
tion, which is in fact false, such that if the subject came to learn
that it was false he would no longer take himself to have good reason
to believe that p and would accordingly drop the belief. And if
this be found objectionable on the grounds that it idealises the rationa-
lity of the subjectwho is required first to see that the new evidence
leaves him without good reason for p, and then having seen that

See G. Harman, pp. 479 andpassim, and K. Lehrer. It is interesting to compare
F. P. Ramsey.

to stop believing pwe can reformulate again in more objective

terms: if he learnt that the said falsehood was false he would no
longer be justified in believing that/). (With this we are very close
to the version favoured by Lehrer.) 3
Now it is clear enough how the Gettier examples trade upon
the existence of the false belief to make it only accidentalin spite
of his excellent reasonsthat the subject ends up -with a truth. Good
reasons, as I have said, will at best only preserve truth; once it
is lost they have no power of their own to regenerate it out of false-
hood, and only the lucky concurrence of circumstances can do it
for them. So one can see the role of falsehood here, and the sense
behind ruling it out if one wants to arrive at logically sufficient
conditions for knowledge. That makes it sound as if the 'No false
lemma' principle must be a necessary condition for knowledge; but
the matter is not so simplewe shall return to it. For the moment
we may concentrate on the question whether it is sufficient, or more
precisely: If we add it as reinforcement to justified true belief, do
we then reach a set of conditions that are logically sufficient for
'S knows that />'? Is there any other way in which the subject's
belief that/), whilst true, could be true only accidentally?
I think that the following, at least, can be said: if we know of
a potential informant both that his belief as to whether p is based
on good reasons and that he satisfies the NFL condition, then we
may deduce that it is very probable that he is right in his belief.
If all we knew were just that he has good reasons for believing
that p (or that not-/)), then we could not deduce any such thing,
because there might be some further truth such that if it were added
to our evidence we would no longer regard his being right as very
likely at all. It should be clear, however, that this (the addition
of something else to our evidence) is the only way in which our
estimate of the probability can be lowered, given the idealisation
that we are operating rationally. (There is nothing remarkable about
this exhaustivenessif we are operating rationally we will not change
the estimate unless there is a change in the evidence.) But coming
to believe a further truth is ipso facto coming to realise that some
falsehood is falsethe falsehood, namely, which the negation of
that very truth expresses. So if -we say, by imposing the No-false-
lemma requirement, that there is no such falsehood, then we in
K. Lehrer, p. 174.
S E C T . IX 79

effect say that there is no truth competent to disturb the original

estimate of probability based on the fact that the subject has good
reasons, which must therefore remain high.
It appears, then, that the NFL condition is immensely powerful.
It would be very surprising if its addition to the analysis of knowledge
as justified true beliefor to any otherdid not result in sufficient
conditions for knowledge. For if we take any condition X such
that on the evidence that S has X we would regard it as highly
probable that S's belief as to whether p is true, and then add to
X that there is no false lemma, this guarantees that the probability
will remain high whatever new facts come to light. And that, given
our core hypothesis, ought to be sufficient for the situation to strike
us as an intuitively satisfactory case of knowledge.
Of course, the very power of the NFL condition exposes it to
a threat from the other direction. So broad is its blanket that no
inquirer will ever have it in his grasp. Along with recognising the
property X of the potential informant on which he rightly makes
his original high estimate of the chances of getting to hear the truth,
he also has to know that there is no further truth such that, going
on the conjunction of it with X, he would (again rightly) significantly
reduce the estimated value. That calls for a kind of near-omniscience
on the inquirer's part, and the call has two obvious disadvantages.
First, we can be quite sure that it will never in practice be answered:
no inquirer ever has that much knowledge. Secondly, if some imagin-
ary inquirer did have it, and if it played an essential part in his
decision to accept that person as likely to offer the truth as to p,
his approach to the 'informant' would look much more like the
approach to a source of information. 4
And if the NFL condition is so powerful, what of our earlier
impression that it is a necessary condition for knowledge? That
impression was based on the argument that, if the subject acquires
a true belief as the conclusion of a chain of reasoning which at some
point involves a false premise, then the truth of his belief is accidental:
reasoning has power to preserve the likelihood of truth, but not
to restore it.
Now this argument does seem to me to hold if we are thinking
of the categorical version of the NFL principle, that is to say the
version according to which the subject must not have engaged in

See Section V.

any actual process of reasoning that made essential use of a falsehood.

But we have observed that its proponents are, understandably, not
very keen on this version of the principle, for reasons indicated
at the beginning of this section. More substantively, adding the NFL
principle in this form to the standard JTB analysis does not yield
sufficient conditions. One reason for that is that it may be satisfied
trivially in those cases in which the belief arises without mediation
of any process of reasoning, as in perceptionwhere nevertheless
Gettier-like examples can still be composed. But another is that
we may imagine our subject to reach the belief that/; by reasoning
that makes perfectly acceptable use of a very restricted range of prem-
ises, negligently giving not a thought to a number of other facts
with which he is very familiar. Had he taken them into account
he would never have come to believe that/), since taken in its entirety
his evidence speaks strongly against it. But, freakishly, his total
corpus of relevant evidence is misleadingp is in fact true, for reasons
of which he knows nothing. Not many will want to say that he
knows that/?; but he certainly satisfies the categorical version of
the No-false-lemma condition.
We turn therefore to the counterfactual formulation of the princi-
ple. It may give us sufficient conditionsas I have argued, we should
certainly expect it to do sobut is it, so to speak, really necessary?
I very much doubt it. I would expect there to be plenty of cases
in which we would feel at least strongly inclined to ascribe knowledge
when it is not satisfied. The basic recipe is as follows:
Begin by letting S have very strong evidence (call it all ) for
p, and let him believe that p on the basis of this evidence. We may
imagine it to be as strong as we like, provided it does not actually
entail the truth of/?. Now let there be some other truth, D, such
that on the basis of ( & D) S would rate p much less likely. D
has always been a very good guide to the truth of not-/? (though
again, not a deductively valid one), and had S come to believe it
(in addition to all of E) he would have fallen into complete confusion
as to what to think about p. But on this occasion the discovery
of D would have been merely misleading: in the unique circum-
stances now obtaining it lacks its usual connections with not-/?.
Examples cooked to this recipe fail the counterfactual form of
the NFL principle: not-D is the false lemma in question. But in
many such examples it will feel perfectly natural to say that the
subject knows that p, even though he would not have known it

(or even believed it) had he had the sheer bad luck to have the
totally misleading matter of D brought to his attentionand brought
to his attention by itself, so that he was not in a position to judge
that D was, just this once, nothing but a red herring.
In summary: the No false lemma principle may be taken in a
form in which it is indeed necessary, but then adding it does not
yield sufficient conditions. Or it may be taken in a form in which
it combines with JTB (and, I suggest, with pretty well anything
else ever seriously advanced) to give a set of sufficient conditions;
but in that form it is not necessary. The conjecture derived from
our 'practical explication', that one will achieve sufficiency only
by including conditions too strong to be necessary, survives this

Certain cases noticed earlier threaten trouble for our hypothesis.'

There was Luigi, who knows exactly what happened to Mario,
but is no use as an informanthe's not telling. There was Matilda,
who knows (this time) that the house is on nre, but is no use any
more as an informant because nobody believes a word she says.
There is the secretly studious milkman, who actually knows the
answer to the abstruse question that is bothering you, but is no
good as an informant because nothing about him gives the slightest
hint that it would be anything but stupid to ask him. (Even his
best friends were never aware that he once swotted the topic up
when a reserve for a heat of Mastermind.) Remember also Colin
Radford's French Canadian: who would have guessed, before they
chanced to start on that game, that there was the slightest point
in asking him even the simplest question about British history?
In all these cases a gap seems to open up between our natural
ascriptions of knowledge and the concept we have arrived at by
considering the practical situation of the inquirer seeking an inform-
ant. To come to grips with them, we need to step back from the
concept of knowledge, and from that of a good informant as well,
in order to look at a far more general principle very widely involved
m concept-formation. It is what I shall call the principle of objectivi-
sation, and it will prove fecund enough not only to help us under-
stand the types of case just mentioned but also to offer insight into
a phenomenon at first sight quite unrelated to them: the existence
and durability of philosophical scepticism.
We start off in terms as general as possible. A creature has a certain
need and desires its satisfaction. What it wants is something "which,
there and then, will satisfy the need; and unless it is the sort of
thing which just comes completely unbidden, the creature must be
able to register it as such so as to be able to orientate its behaviour
See Section II, last para.
SECT. X 83

towards it, and it must have the necessary motor-capacities to do

so. It needs this situation as a wholeif it all obtains, the creature
succeeds, if any of it is missing, the creature fails and the need persists.
So if all we are thinking of is the fulfilment of this need here and
now, it has no cause to distinguish the various aspects: the presence
of food, its own capacity to be nourished by that food, its own
capacity to detect the food and reach it. Thus, perhaps, the barnacle.
But with the slightest hint of intelligence this primitive holism starts
to fragment. The creature must distinguish between food, here, now,
provided it makes the right movement, and food here, soon, pro-
vided it waits very quietly for a bit and then makes the right move-
ment. It must distinguish these from food, there, soon, provided
it can get there; and cases in which it can get there from ones in
which it cannot. Helpful again, as life grows more varied, will be
the capacity to distinguish cases in which it simply cannot get there
from those in which it cannot get there because of some temporary
hindrance, either in the environment or m itself. These are differences
in the situation which require different strategies, and the creature
which can respond to each of them appropriately will be more likely
to prosper.
That creature was an individualist. If we place it in a social group
the possibilities expand again. What it cannot reach perhaps a
maturer individual canwhich suggests an all too familiar way of
getting at food that would otherwise be inaccessible: point it out
to Mum. Others may be better than it at recognising food; so far
as they seem trustworthy, follow their recommendations. Others
may be able to use substances as food which it cannot; see that
they get it. (That will be encouraged by the existence of natural
altruism, if there be any, but it can and will arise without it: it
isn't necessarily for the sake of the horse that we give it the oats
instead of eating them ourselves.)
What we see here are the natural pressures driving thought away
from the totally subjectivist stance, the pure 'here and now for me
as I am here and now' that I slanderously imputed to the barnacle.
They induce the thought of the satisfaction of a need at other times
and other placeswhether that be the present need or a similar
need expected for the future, the thought of recognitional and beha-
vioural (and maybe digestive) capacities other than those I have here
and now, and hence of an object which can in the right circumstances
satisfy such needs whilst, for any number of reasons, coming
84 SECT. X

nowhere near to meeting the wholly subjectivised conditions from

which our thought-experiment began.
One more everyday example, before we turn to informants. I
may well be interested in 'something which I can now sit on' (only
close and accessible objects need apply). But in due course I shall
be interested, since I anticipate wanting to sit down at future times,
in objects which I could sit on if I wanted to, or in whether there
will be something which I can sit on when I want to (at the end
of the walk). This interest will naturally lead to an interest in hearing
the opinions of others as to where there are objects which I can
sit on if I want to, irrespective of whether they want to sit on them
or not; so I shall want them to operate an objectivised concept
too. And if I grow a little more altruistic in my outlook I may
even be interested in whether there is something which Fred can
sit on if he wants to, irrespective of whether I shall want to sit
on anything or not. Hence the concept of something which is, in
abstraction from what any particular person wants at any particular
time or place, or even from whether anyone ever wants to sit down,
simply suitable for sitting on. It may right now be out of reach,
it may be upside down, it may be folded up in a cardboard box,
perhaps no-one will ever want to sit on it anyway; but it is a chair.
I oversimplify, of course; there is more to the concept of a chair
than that. Where a particular type of physical object is created to
meet a need, physical characteristics as well as fitness for a certain
purpose enter into the classification. But the simplification does
not affect the point of the example, which is to illustrate the idea
of the objectivisation of a concept, and at the same time to explain
why we have objectivised concepts.
The explanation, we may note, is still of the 'state of nature' var-
iety. It need not presuppose that the wholly egocentric, 'subjecti-
vised' thought from which it began actually exists or existed, any
more than a Hobbesian account of the State needs the corresponding
presupposition about the war of every man against every man. The
argument is only that if it exists, at any time, or in any individual,
it will develop in the direction of objectivisation. Therefore there
will be objectivised concepts, whether things started that way or
We can now apply these ideas to the situation of the inquirer
and the concept of a good informant. We begin by considering it
at its most subjective. I am seeking information as to whether or
SECT. X 85

not p, and hence want an informant who is satisfactory for my pur-

poses, here and now, with my present beliefs and capacities for
receiving information. I am concerned, in other words, that as well
as his having the right answer to my question,

(1) He should be accessible to me here and now.

(2) He should be recognisable by me as someone likely to be
right aboutp.
(3) He should be as likely to be right about p as my concerns
(4) Channels of communication between him and me should be

There are various ways in which a candidate might fail these con-
ditions. As regards the first, he might just not be here when I want
him. Notice that in this context 'here' does not quite mean 'in the
same place as I am'; it stretches to include all places which our
systems of communication link with the place where I am. But at
this stage we must also build in 'or soon can be'for another subjec-
tively varying factor will be the urgency of my situation; the less
urgent, the laxer we can afford to be over the spatio-temporal
Equally, there are various ways of failing condition (2). One of
them, of course, is by simply not being likely to be right about
p. Another, if it is possible (I shall not try to resist those who feel
that it isn't), is this: whilst being likely to be right about/), possessing
no property whatever from which anyone, however knowledgeable
and acute, could infer that one was likely to be right about it. Neither
of these, clearly, have anything in particular to do with the state
of our particular inquirerif they hold, they hold for every
inquirerbut there are many that do. For supposing that the candi-
date does possess a property which correlates well with being right
about p, inquirers may (and mostly will) differ in point of their
capacity to detect the property, and in point of knowledge of the
correlation. One needs both of these to be able to detect the potential
of the potential informant, so some inquirers will be better at it,
some worse. Such differences may depend on sensory acuity, intel-
lectual facility, theoretical background; they may also depend on
altogether more transient features of inquirers, like their spatial pos-
ition or bodily orientation. But all introduce a subjective element
86 SECT. X

into the question: is Fred a good informant, for me, now, as to

Factor (3) gives a further dimension of subjectivity. The informant
should be 'as likely to be right about p as my concerns require'.
Take the inquirer, looking for a good informant as to whether p,
in certain specific circumstances for certain specific purposes. We
have seen that he will be interested in the candidate's competence
not just in the actual world, but also in various possible worlds.
Not, however, all worlds which strike him as theoretically possible,
but only a selection of these. Some he will exclude because he has
already come to believe that, although possible and even antece-
dently quite probable, they are not actual: he can see that the poten-
tial informant is wearing a red shirt, and so excludes those worlds
m which he is wearing a blue shirt as merely possible and non-actual.
In our earlier terminology, they are not 'open' possibilities for him
any more.
Other possibilities he will exclude on slightly different grounds.
Without actually taking them to be false he will regard them as
so improbable that they need not be taken into account. So if the
candidate would fail (hold the wrong belief as to whether p) in some
world W, he will not allow that to influence his choice of informant
if he puts the likelihood of Ws being the actual world low enough.
How low? That will depend on a number of factors. One is the
urgency of forming a belief as to whether/?, in the inquirer's particu-
lar situation: sometimes the penalty for not forming a belief at all
(and so just dithering), is as great or greater than that for forming
a false one. If my tram leaves in 5 minutes I had better come to
a decision about whether the station is this way or that way. Guessing
will be better than not deciding, and anyone with a better than
evens chance of telling me the right answer will be welcome. In
such a storm, almost any port. (Had Descartes started on his medi-
tations without first taking thought for his immediate practical needs,
he could not have afforded to suspend virtually all belief on the
chance of a malm genieas he was well aware.) Another case con-
cerns the relative pay-offs of being right and being wrong: if being
wrong won't matter too much, but being right will be very advan-
tageous, I may be satisfied with an informant of lower reliability,
one whose views have a lower probability of being true, than I
would be if the situation were reversed, so that being wrong would
turn out very damaging. (The limiting case is surely Pascal's wager,
SECT. X 87

where we are asked to adopt a belief without learning anything that

raises the chances of its being true by any amount at all).2 A further
element is my personal attitude to risk. I may positively enjoy it,
and so be prepared to take risks which I would not take if I enjoyed
it less; one type of risk-taking is being somewhat less demanding
in one's choice of informant. Or the converse: my natural inclination
may be towards security. And all these calculations must be seen
against the background of my circumstances. The rich aren't inter-
ested in winning small bets; for the poor a small stake can represent
a big risk.
Lastly, there are many factors affecting fulfilment of condition
(4), lying in various aspects of the relationship between inquirer
and (potential) informant, and so liable to alter if either changes.
Lack of a common language is onefor some values of/), though
by no means all, it can often be overcome, by the use of gesture,
for instance. Another is unwillingness to part with the information,
something which, for a given informant, characteristically differs
from one inquirer to another, and perhaps for the same inquirer
from one time to another; that was the position of our gangster
Luigi. Another concerns credibilityit can be that someone who
clearly has an excellent chance of having a true belief about p is
disqualified as an informant because there are reasons to doubt his
sincerity; such was the position of Matilda, who told such dreadful
lies that in the end no-one would believe her even when she was
telling the truthand this even though they could tell that her belief
on the (may I say it?) burning question was almost certain to be
the right one.
Whether I (subjectively) rate Fred a good informant depends in
all these ways, and possibly a few more, on my situation and my
relationship (in the broadest sense) to him. In pursuing my ends
I must take them all into account. Others, who may be interested
in Fred as a possible source of information, will not; they will have
their own versions, arising out of their situation and their relation-
ship to him.
What we have at this stage, then, is a number of individuals with
the same problemhow to come by the truth as to whether pand
their various ways of approaching it, determined by their individual
requirements and circumstances. But these individuals form a com-
B. Pascal, para. 418.
85 SECT. X

mimity, and are in some degree at least helpful to others and respon-
sive to their needs. And even if I, as one of the community, am
not so inclined, I shall still need an appreciation of their point of
view if I am to be any good at getting them to help me. From
such facts arises a pressure towards the formation of 'objectivised'
concepts, concepts which separate as it were the common core from
the multitude of accretions due to particular circumstances and parti-
cular persons and so varying with them.
To see how this may come about, consider that I shall hope that
others will on occasion recommend informants to me. (This includes
the case in which someone recommends himself.) I am not, of course,
asking them to recommend people who, without their recommenda-
tion, would in any case have struck me as good informants. (This
is one way in which my appreciation of their point of view comes
in: if I am to be any judge of whom to ask for a recommendation,
I shall need some idea of what sort of person might look like a
good informant to them.) Quite the contrary: I want them to recom-
mend as informants persons whom but for their help I could not
have recognised. There are many detailed reasons why others may
be in a position to do me such a service, but in general they come
down to this: that they can detect properties of the informant which
I cannot detect, or have more knowledge than I have of which
properties correlate well with being right on the topic at issue. In
practice this will mostly be a matter of their being 'better placed'
than I am. As I seek to discover the close-of-play score there may
be nothing about Fred to suggest to me that he would be a good
person to ask. But Fred's friends are aware that he was at the ground
when play finished; so their advice will help me. Amongst the best-
placed observers of Fred is of course Fred himselfwhich is why
people are so often in a good position, indeed often the best position,
to advertise themselves as good informants.
As well as properties of Fred which I adventitiously cannot detect,
there may be some which I am in general poor at detecting, and
in the event of any of them correlating closely with Fred's capacity
to form true beliefs on whatever it is I am curious about, I will
hope that others, better endowed, will be prepared to point me
towards Fred. In either case I am constructing the idea of someone
who is a good informant, but whom for one reason or another I
cannot detect as such. And I can go further, and conceive of someone
who has a property correlating excellently with possession of the
SECT. X 89

truth as to whether/;, but which property no member of my com-

munity, or no member of my species, can detect.
Very similar considerations as those which apply to the correlating
property also apply to the fact of the correlation. There may be
many cases in which I can recognise the property X without diffi-
culty, but not that possession of it correlates well with having the
truth about p. Others may be better placed, and I shall hope for
their goodwill and assistance. Again I can go further, and conceive
of the case in which grasp of the correlation called for theoretical
knowledge which nobody had, hence of somebody who might be
discovered to have been a good informant but whom nobody could
at present recognise as such. Perhaps I may go further still, and
conceive of a correlation the recognition of which is not just beyond
present human knowledge, but beyond human powers.
Some readers may feel that this kind of talk gives insufficient
weight to verificationist lines of thought. But they shouldn't worry;
nothing I have said here presupposes anything about venficationism,
for or against. The suggestion was that someone might know some-
thing when we could not tell that he knew it. The verificationist
warns us, on pain of insignificance, not to let that 'could not' become
too strong. Let us agree, though without agreeing on any particular
way of specifying the limits, and assign to 'could not' the maximum
legitimate strength, whatever it may be. All I am trying to explain
is how the process of 'objectivisation' of concepts might lead us
to that thought, whatever it is.
It has to be said however that there is a good reason for the verifica-
tionist's discomfort that has nothing to do with venficatiomsm. We
are attempting a 'state of nature' explanation of a number of facts
of conceptual or linguistic practice. Such explanations work by iden-
tifying certain human needs and arguing that the practices are a
necessary (or at the least a highly appropriate) response to them;
they will therefore be at their strongest when the human needs from
which they start are the most practical, hence the most undeniable
ones. This sets limits to what a 'state of nature' explanation can
be good for. The less visible the practical significance, for us, of
forming a certain conception or operating a certain linguistic usage,
the weaker the explanation. Either it will become harder to mount
a convincing argument that the needs we start from really do require
the explanandum, or the less the needs will look like uncontroversial
facts of the human condition. And if the visible practical significance
90 SECT. X

sinks to zero, as one might think it had when we form the conception
of a state of affairs which we are totally incapable of detecting, the
force of a 'state of nature' explanation will sink to zero too, unless
it takes the liberty of helping itself to additional explanatory princi-
ples. This is not to say, with the verificationist, that there is no
such conception, only that a certain style of explanation is impotent
to account for its existence, if it does exist.
But since for the moment we need only concern ourselves with
covert and not with strictly undetectable properties we don't need
to tread on any verificationist toes, or (what would worry me more)
to go beyond the effective range of our method in order to use
these thoughts about objectivisation for clarifying some of the prob-
lem cases. They indicate sufficiently why we will come to distinguish
between the properties of the object which make it suitable for a
certain use, and the properties of the user which make him capable
of using it. To put it briefly: the two facts that human life is social,
and that the members of a human society have differing capacities
and perspectives, make it obligatory to form the separate conceptions
of the state of the object, which is invariant with respect to different
individuals, and the states of the 'consumers' of the object, which
are not. We are in the same area here as are those ethical theorists
who point out the social advantages of adopting moral principles
which prescind from facts specific to individuals.
What happens to the concept at the centre of our investigation,
that of the good informant, as objectivisation proceeds? The require-
ment of a true belief remains, and so does that of a property correlat-
ing well with truth of belief on the issue in hand, but that of the
detectability of the property will be diluted. When we began the
individual inquirer was looking for something that he could recog-
nise there and then, preferably with a minimum of effort. That,
of course, is what he will still hope for, but it is not what will
be embodied in the public concept that now develops. For one thing,
what is effortlessly available to him then and there will not be a
matter of public interest, nor will it be the only thing that is of
interest to him; for another, it will be in his interest to bear in
mind the possibility of others having different (which will include
more powerful) powers of detection, since that will at times be of
use to him, let alone to them. The more we get used to the existence
of such powers, the weaker will the detectability requirement, as
reflected in the public concept, become. The concept of knowing,
SECT. X 91

our hypothesis must now run, lies at the objectivised end of the
process; we can explain why there is such an end, and why it should
be found worth marking in language.
Earlier in this section we looked also at the third factor and the
determinants of its most subjective versions, so to speak its 'me-here-
now' shapes. How important is it to the particular inquirer, in his
particular circumstances, to get the truth? What is his attitude to-
wards risk? These are equally ripe for objectivisation. Just as I may
have an interest in being able to sit down where and when I want
to, without knowing exactly where or when that will be, so I may
have an interest in collecting information while it is available, without
knowing when, or why, or under what pressures, it may be needed.
In addition, I shall be concernedwithout knowing in detail what
their circumstances and purposes arethat others should make
judgements as to who is a good informant. That may be because
I have enough altruism to wish that their enterprises in general,
hence also their enterprises of belief-acquisition in particular, will
succeed. But even without it I shall still want them to make assess-
ments of informants, because that may turn out to be useful to me.
I shall not suppose, however, that in making such assessments they
have my particular circumstances in mind: if they tell me whom
to ask about the times of trains to London, I shall not expect them
to take into account how important it is to me to get the right
answer, what I shall lose if I don't arrive on time. Perhaps I don't
yet know that myself; perhaps it isn't even on my own behalf that
I am trying to find out, but for someone else whose exact concern
with the information is unknown to me.
All this is going to edge us towards the idea of someone who
is a good informant as to whether p whatever the particular circum-
stances of the inquirer, whatever rewards and penalties hang over
him and whatever his attitude to them. That means someone with
a very high degree of reliability, someone who is very likely to
be rightfor he must be acceptable even to a very demanding
inquirer. So of the worlds that we cannot quite definitely exclude,
we shall want to include in our assessment of him even those that
we regard as very improbable. Moreover, we shall be motivated
to take a pretty careful look at those which we 'can quite definitely
exclude'is that really as many as we think? These thoughts take
us further down the road of objectivisation. Knowledge, so the
hypothesis goes, lies at the end of it.
92 SECT. X

My interest in other persons' powers of information-collection

does not just arise from the fact that I may sometimes wish them
to pass it on, or recommend informants to me. I shall become
involved in group action, and affected by the actions of others; then
circumstances will arise in which it is important to me that someone
in the group holds a true belief as to whether/;, and quite unimpor-
tant whether the route by which they acquired it would have been
open to me or not. (I am very pleased that there are people who
know how to disarm and dismantle a nuclear missile; but whether
those who instructed them would also be prepared to instruct me
is of no interest to me at all.) This throws light on cases in which
condition (4) fails: that is to say in which a person is no good as
an informant because the channels of communication are for some
reason not open. Luigi and Matilda are both said to know, although
hopeless as informants; he won't say, she won't be believed.
One way out of this difficulty would be to switch the hypothesis
to the first-person case. If what we are concerned with are the cir-
cumstances under which we take ourselves as informants, so to
speakwhen we decide that one of our own beliefs is reliablethen
problems about channels of communication will not enter into it,
since they may be assumed always to be open. But although there
may be something in this (I certainly don't want to deny that first-
personal considerations may have a role to play in shaping the con-
cept) it would be feeble to make it the mam plank of our explanation.
Concern about the truth of one's own beliefs calls for some theoreti-
cal sophistication, and although I would hazard a guessit is not
so very hazardous, when one comes to think of itthat it is a degree
of sophistication that exists wherever language, and the concept of
knowledge, are found, it is still preferable to keep the basis of our
investigation as primitive as possible, which means sticking to the
third-person approach unless absolutely forced to abandon it.
Whether that point will ever come or not, we haven't reached
it yet. Just as properties of the potential informant are not simply
detectable or undetectable, but detectable for some under certain
circumstances and undetectable for others, or under other circum-
stances, so the question whether channels of communication are
open or not isn't a simple Yes or No matter, but depends on facts
about the particular inquirer and his relationship to the possible
informant. Luigi won't tell me what happened to Mario, but he
may tell Carlo. Now the objection may be made here that unless
SECT. X 93

Carlo is prepared to spill the beans (which he isn't, unless Luigi

is a very bad judge of whom it is safe to talk to), that won't make
much difference to meso we still don't have an account of why
I should be motivated to form the objectivised concept which applies
to Luigi although he won't tell me. True, but not all examples are
like that. My student won't tell me what is worrying him, but per-
haps he will tell someone at the university counselling service, and
then (whether the news is passed on to me or not) something may
get done that will helpand incidentally that will make my life
easier as well as his. It will surely be to my disadvantage if I don't
have the conceptual resources to recognise the visiting Chinese dele-
gates as a source of information just because there is a blockage
in the linguistic pipeline. After all, they will happily tell the inter-
preter, and the interpreter will tell me.
Some philosophers, on thinking of this type of prima facie counter-
example to the hypothesis that knowing is being a good informant,
will say at once that knowing seems to be that central state of the
subject which fits him to be a good informant, other circumstances
concurring. It isn't, on my view, that they are wrong. My criticism,
if that be the right word, is just that it would be more satisfying
to have some explanation of how and why such a concept comes
to be formed. That is what the idea of objectivisation provides. But
there is another point to be noticed: the concept thus explained
is not quite captured by the above form of words. The words were,
'that central state of the subject which fits him to be a good informant,
other circumstances concurring', and one natural way of reading
them is a long way from the concept of knowledge.
We have seen that, if other circumstances concur, someone with
only a fairly good chance of being right may be perfectly acceptable
as an informant. Perhaps no-one more likely to be right exists, the
prize for holding (and following) the right opinion is high, the
penalty for following the wrong one is trivial. But that informant
is not rated as knowing. For that an informant needs a very good
chance of being right, so that the concept we are looking for is
more like that of being in a state which makes him a good informant
about p whether other circumstances concur or not. Such is the pos-
ition when we consider condition (3), but we find just the reverse if
we focus on condition (2). In respect of condition (2) nobody is ever
a good informant unless circumstances concurif they don't, even
the most obvious of correlating properties may be unrecognisable,
94 SECT. X

or unrecognisable as such, and the information-potential go

Is this asymmetry damaging to our thesis? I think not. The thesis
brings with it no commitment to the view that the objectivisation
of each of the features (1) to (4) (plus any others there may be)
proceeds in the same way, or in such a way that some simple verbal
formula can be found that neatly fits them all. It demands only
that what happens in each case be readily understandable in terms
of motivations which our hypothesis suggests; and about this there
is no great difficulty. In saying that someone knows whether p we
are certifying him as an informant on that question, and we have
no idea of the practical needs of the many people who may want
to take him up on it; hence a practice develops of setting the standard
very high, so that whatever turns, for them, on getting the truth
about/), we need not fear reproach if they follow our recommenda-
tion. (Where, as in the murder trial, we positively know that the
most serious consequences turn on it, our inclination is to wind
the standard up yet another notch.) Condition (2), the accessibility
of the correlating property, plays a very different role. In recom-
mending an informant to you I am indeed implying that the likeli-
hood of his being right is as great as your concerns require; but
I am not implying that the facts that signal that likelihood are access-
ible to your cognitive equipment. Indeed, it is just when they are
not that you most stand in need of my recommendation.
The point turns on a lst/3rd person asymmetry. The inquirer
can make use of another's identification of the good informant, but
he must act on the information, when it comes, himself. So when
I identify a good informant on your behalf I need not bother myself
with the question whether you could, in principle, have made the
identification yourselfit just doesn't matter. But I do have to
bother whether the reliability of the informant meets your standards,
since it is you who will act on the information; therefore, if I don't
know what you want it for or what fate awaits you if you get it
wrong, I had better make sure that the level of reliability is very
high indeed.
Some readers, I anticipate, will have got the feeling that the cart
is before the horse. Don't you need to know first, before qualifying
as an informant? Shouldn't being a good informant be explained
in terms of knowing whether/;, rather than the other way round?
The reaction is obviously a natural one, and more than that, there
SECT. X 95

is a good deal of force in it.

We are going to have to accept that the relationship, if it exists,
is between knowing and having potential as an informantwe have
seen that someone may be a perfectly good knower when, for one
reason or another, the channels of communication are closed or
malfunctioning. And this may be taken to suggest that being a good
informant is knowing plus some other factor (communication, credi-
bility); and this in turn to mean that knowing is primary, being
a good informant derivative, contrary to my approach.
Nevertheless there are, compatible with this admission, two ways
of looking at the relationship between knowledge and capacity qua
informant. One would be to say that the former is in fact best under-
stood without reference to the notion of value as an informant, and
that the latter then arises out of it as a welcome social bonus. That
line is indeed incompatible with my project, but there is another
which is not: that knowing is best understood in terms of the idea
of a potential informant, that we get at its nature by first considering
the business of being a good informant and then subtracting some-
thing from the result, or modifying it in some way. The concept
of the informant will then leave its mark, so to speak, on the concept
of knowledge, and the hypothesis is that for this reason an approach
via the former concept will be likely to provide a more illuminating
account of the concept of knowledge than will any other method.
It may improve the reader's grasp of this point, as well as perhaps
raising his estimation of its prospects, to return to an example which
has served us earlier in this section. Consider the imaginary case
of someone interested in the concept of a chair, who observes that
most of us like to sit down, and suggests that the concept of a chair
arises as part of our response to this need; this, he urges, is likely
to prove a particularly illuminating approach. Thereupon a dissent-
ing voice is heard, warning us that we are in danger of getting the
cart before the horse. Surely we can see (thus the voice) that things
that are not in position to satisfy the need to sit down can perfectly
well be chairs. If I need to sit down, only something which is then
within my reach can satisfy the need; a chair dropped off by some
expedition in the middle of the Sahara can't satisfy anyone's need
(there's no one within 200 miles of it, let us say); something might
be a chair although it could only be used by a young child; a chair
on top of a bonfire that is already blazing is still a chair. So (still
the dissenting voice speaking) don't try to explain what a chair is
96 SECT. X

by reference to the satisfaction of the need to sit; first tell us what

a chair is, then it will be clear that, and why, chairs are good for
sitting onprovided they are in the right place, and of the right
size, and the right stability, relative to the prospective sitter.
I offer this as a prima facie parallel to the objection that, since
one can easily think of cases in which a knower may not be a good
informant (because something or other is blocking the channel of
communication) we should not give priority to the latter idea in
giving an account of the former. If it is parallel, then obviously
hope remains; for most will agree that the dissenting voice of the
previous paragraph has got things wrong somehow; and if it has,
then maybe the comparable objection in the case of the concept
of knowledge is in the same boat. One would not find out much
about why there are chairs, or why the concept of a chair has the
boundaries (fuzzy though they be) that it has, by concentrating
attention on the shape and constitution of the things to the exclusion
of what they were for. To be lead into that state of mind by the
observation that some chairs, for one reason or another, cannot
be sat on, would be a heuristic disaster. A chance has to be given
to the hypothesis that the same is true of knowledge, and the objec-
tion equally misconceived.
To show that it certainly is misconceived would be to show that
the project of this essay, indeed the particular hypothesis under
consideration, certainly succeeds, and that may be asking too much.
But we can fairly quickly say enough to show that we are perfectly
justified in proceeding, and that the 'cart before the horse' objection
shouldn't be allowed to bring the investigation to a halt. In the
first place, we imagine, the members of any society will be interested
in obtaining information from their fellows, and to this end they
will develop a concept very like our concept of a good informant
one who (predictably) believes the truth on a given matter, and
is prepared, and available, to pass his opinion on comprehensibly
and without dissembling. The process of the objectivisation of con-
cepts which may be thought of as having started out in much more
heavily subjective versions has already been described in outline.
It will lead to a concept which so far as possible leaves out of account
those features of the original concept that fitted it for the particular
use of the person applying it, with his particular location, powers
of discernment and comprehension, relationship to the potential
informant, and so on. What remains will not be identifiable with
SECT. X 97

the concept of a good informantas indeed we have already seen

that there are many know^rs who for one reason or another are
not good informants; but it may still bear certain marks of its origin,
and if it does then the concept of the good informant may be the
key for drawing them to our attention and making their presence
comprehensible to us. It will also bear the marks of the objectivisa-
tion. Some of the marks may obscure some of the others.
By envisaging such a story we do several things which the propo-
nent of the 'cart before the horse' objection does not. If we say:
'Knowledge first, and then comes the capacity to inform', we leave
it obscure why the concept of knowledge should ever have arisen,
rather as the comparable objection about the concept of a chair would
leave that concept, and even more so the chairs themselves, unac-
countable brute-fact bits of the mental and physical habitats. Of
course the objector may (in the case of knowledge, scarcely in that
of chairs) go on to make a positive contribution in the shape of
some alternative hypothesis. Then we havewhat the cart-before-
the-horse objection by itself does not give usan honest confron-
tation, which is to be decided, if at all, in the way in which such
confrontations are usually and properly decided: by weighing the
plausibility of the opposing views, the amount of detailed fact which
each can explain, the elegance and accuracy with which it explains
Taken on its own without the support of any alternative hypothe-
sis the objection does not therefore have much force. Nor does it
suggest any further lines of inquiry or explanatory possibilities,
whereas this is precisely what my theory does. As well as offering
a plausible account of the purpose of the concept, it introduces the
notions of the good informant, and of conceptual objectivisation,
and these may have other uses besides those we have so far seen.
Objectivisation, for instance, may help us understand the phenome-
non of scepticism. But that is a large topic, and deserves a section
of its own.

In the preceding section one of the questions we considered was

how the third condition (that the candidate-informant be 'as likely
to be right about/? as my concerns require') will fare as objectivisation
proceeds. What we saw was that there will be a shift towards adopt-
ing a high value for the likelihood required before we are willing
to recommend an informant. For the conditions under which we
recommend informants will not, in general, be ones in which the
recommender is sufficiently aware of the concerns for which the
informant and his information are needed. This fact will push the
required standard up to such a level that the recommender may
responsibly issue his recommendation whilst knowing nothing of
them, that is to say, to a level which the recommender may reasona-
bly take to be high enough to satisfy all, or all practical, purposes.
Our use of 'knows', so says the hypothesis, marks the attainment
of that level. On the other hand, it would be most disadvantageous
if we were unwilling to offer any information except when we took
that level to be reached; so for lower levels we have more guarded
forms of words available: 'I'm pretty sure that ...', 'He's usually
a good judge of this sort of thing', and many others.
Is there anything more we can say as to just what level this is?
The above line of argument would suggest a fairly stark answer:
the probability that the belief is true needs to be 1, no less. For
if it is less that 1, can we not imagine an inquirer whose concerns,
whose game-theoretic situation, are such that that likelihood of truth
is insufficient for him rationally to commit himself, and consequently
also insufficient for another responsibly to commit himself on his
Important here, however, is the word 'imagine'. As the prob-
ability approaches 1 it becomes progressively easier to imagine such
an inquirer than to identify one. And what one encounters only
in the imaginationif indeed one's imagination runs to it at allhas
very much less influence in determining the shape of an everyday

conceptual practice than what a community actually encounters in

its living experience. If we are forced to put a precise numerical
value on an acceptable probability-coefficient, it may well be that
it can only be 1. But if we are forced to do that it is we who are
forcing ourselves. The subject matter does not demand it, or even
encourage it; we are talking of an all but universal everyday practice,
and everyday thinking about likelihoods does not go on in such
precise quantitative terms; it is far too like a feeling for that. What
we should do here is look to imprecisely quantified concepts like
'being sure' and 'feeling certain' or (with a eye to Austin's treatment
of the topic)1 to that of offering a guarantee, or promise, of truth.
All of these, as the detail of recent debate bears witness, have a
role in our more or less untutored reactions to questions about the
concept of knowledge, and it is to help us to see why that is so
that these thoughts about objectivisation are appropriately
employednot to generate spuriously mathematicised probability
It is now the moment to break down the fence I may be thought
to have been by implication erecting around 'everyday' practice,
apparently isolating it from non-everyday practices which involve,
and are in part the outcome of, reflective thought. We have no
clear grasp of such a distinction, any more than we have of the
distinction one sometimes hears of between philosophical and non-
or pre-philosophical thinking. But even though the fence is to be
taken down, it has served a methodological point. A recurring theme
has been the universality of the practice we are investigating, and
the way in which this demands an explanation in terms of conditions
which hold just as universally, that is, in every community that
operates a language. That requires, not that we fence off practice
from reflection, since there may be some kinds of reflection which
can with great plausibility be argued to be just as universal, just
as much demanded by the state of nature as it affects beings of human
intelligence, as the particular linguistic or conceptual phenomenon
we are trying to account for; if so, those kinds of reflection have
a perfectly good place in the account. The distinction we should
be aware of is the one between those and the kind of reflection
that is a response to conditions which, although they may be wide-
spread in certain places at certain times in history, or amongst certain
J. L. Austin, pp. 67-9.
100 S E C T . XI

types of people, cannot with the same conviction be thought to

hold with scarcely an exception.
Two such conditions certainly have an effect upon our concept
of knowledge as soon as we begin to reflect on them, and they
are found, I would hazard the guess, in every society that we would
think of as civilised or advanced, as well as many that we wouldn't.
One is the practice of betting on outcomes at quantified rates, so
that potential winnings are roughly proportioned to the judged
unlikelihood of the outcome; a lottery, in which the chances of
winning decrease m exact proportion to the number of entrants,
and the jackpot increases in the same proportion with the increased
number of stakes, is a precisely quantifiable (though doubtless less
widespread) form of such a wager. The second practice is a sophisti-
cated inferential one: that of drawing conclusions that depend upon
a large number of premises.
What the example of a lottery does is to bring much closer to
our real experience the idea of someone who would not necessarily
be well advised to accept information which had a chance less than
1 of being right. Consider the claim that this particular ticket, the
one he is about to buy, will not win. We can make the likelihood
that this claim is correct as high as we like, short of 1, by enlarging
the number of tickets going into the draw. But since by enlarging
the total number of tickets we also proportionately increase the prize
without increasing the stake, it may make perfectly good sense for
our punter, especially if the cost of a ticket is trivial to him, to
reject the information that this one will not win and go ahead and
buy it. Only if he believes the probability to be 1 can it be clear
that he should take the informant to be good enough for his purposes.
The same sort of practical considerations as were at work in the
less precisely quantifiable cases are at work in the case of the large
lottery, and it can be seen how they push our estimate of the prob-
ability that will 'do' up to 1, or certainty. So the idea that someone
only knows whether p if something about the situation makes it
certain that their belief on that question is right has some basis in
the practical, even if it is not a basis that we can expect strictly every-
one to encounter. A similar push is exerted by the wish to draw
inferences from multiple premises and be able to rely on the result.
The point is well-known, as is the fact that the example of the
lottery can be mobilised again to illustrate it. If
I know that a, I know that b, . . . . . I know that n
SECT. XI 101

is to legitimise my claim to know anything that I know to follow

from the conjunction (a & b &C .. . & ), and if this is to hold
for arbitrarily long conjunctions, then I can only claim knowledge
of propositions whose probability I can set at 1. For suppose I
allow that there can be knowledge that p so long as the probability
of p's falsehood is below l/n, I only have to think of a lottery with
more than n tickets to be able to say of each ticket that I know
that it will not win. And then, if we suppose that I also know that
those are all the tickets, the conjunctive principle allows me know-
ledge that no ticket will win, something which, on the contrary,
I actually know to be false.
The wish to be able to rely on the conclusion of an essentially
many-premised inference isn't unique to gamblers, or to citizens
of the scientifically more advanced societies. Consider the chieftain
of a tribe who wonders whether he knows that no member of his
army would betray their plans to the neighbouring enemy. On the
basis of report and personal acquaintance he feels sure, for each
one, that be would not do it. Does he, mindful of his responsibility
for the tribe's security, now conclude that he knows that nobody
would do it, and take no further precautions against treachery? He
will be keenly aware that he only needs to be wrong in one case
out of, perhaps, several hundred. Such situations automatically raise
the stakes, force up the demands on what, for these purposes, is
usable information and an acceptable informant.
Lotteries, of course, have another feature which this situation
does not. They have winners; or at least the participants believe
them to have winners, and would not participate otherwise. So
when we think of the possibility of winning a big lottery we are
not (as we see it) thinking of something that just might happen:
the minute chance of winning, we believe, always comes good for
someone or other. And if we are reflective enough to compare their
epistemic position before the draw with our own and everybody
else's we find, assuming that we believe the draw to have been fair,
no relevant difference. Someone who told them that they would
not win would have been wrong; someone who told us that we
would not win would have been right, but could not have been
Our chieftain is not in quite the same position. He does not
have to face the fact that every army has its enemy agentit
doesn't. But he may well believe, indeed know, that some well-
102 SECT. XI

screened organisations have turned out to harbour traitors, and if

he is wise he must allow this to affect his attitude to his present
What we see from all this is how factors arising out of special,
but not for that reason extremely rare, conditions, given a little
reflection of no very high-brow kind, force up our estimate of the
level of likelihood required. Two things are to be noted at this point:
To find a need to raise the likelihood to 1 takes a little extrapolative
imagination. Real situations sometimes force us to set it very high
indeed, but not actually at 1. Perhaps because of that, the final push
towards a probability of 1 only makes itself felt in circumstances
which invite reflection. In everyday practice we happily bandy the
word 'know' about without having to feel that our chances of being
wrong are literally zeroand this is in good accord with my genetic
account of the concept and pragmatic account of its point.
Secondly, even when we do set the likelihood at 1 that does not
mean the type of absolute certainty, invulnerability to literally any
theoretical possibility, which the sceptic characteristically demands
and finds us unable to provide. Nothing about our reactions to
lotteries or the danger of treachery need push us any further than
the demand that the actual or 'real' chance of being wrong should
be zero. Hence, I suggest, the attractiveness of the idea that someone
knows whether/? when something about him is connected as a matter
of natural law with his holding the right belief about p.2 Such
an analysis, one can now see, has a rightful place in the debate.
But one can also see that mistakes may have been made about 'what
that place is, in particular when it has been announced as the analysis
of the concept of knowledge. What it is, rather, is a reflective tidy-up
of everyday practiceand indeed a well-founded tidy-up, since it
emerges very naturally from the motivation which gave rise to the
practice in the first place. But it also needs to be said that it is,
as tidy-ups go, pretty disruptive. For if we were really to conform
our practice to its recommendations we would have to be either
very cautious, or very trusting, in our ascriptions of knowledge.
Whether we really know enough about the states of subjects and
the laws in accordance with which the mind forms beliefs, ever to
be in a position to say that some belief of Fred's will be true as
a matter of natural law, must be open to serious doubtremember
e.g. D. M. Armstrong, esp. pp. 166 ff.
SECT, xi 103
that we would need to know this before knowing whether the belief
in question was true or not, so a straight appeal to natural determin-
ism (supposing we knew it to hold) is of no avail. Indeed we may
already know enough about the complexity of the brain and the
role which its complexities play in belief-formation to decide here
and now that we will never be able to say that since Fred is in
such-and-such a state his belief as to whether p will be true as a
matter of natural law.

The first step in writing about scepticism must be to say what one
is writing about. A negative attitude towards almost any generally
received truth can be called scepticism, and in the course of the
history of thought a wide variety of positions have been so called.
Certain ancient sceptics were known to recommend abrogation of
all belief in furtherance of inner peace. Hume, for many a paradigm
sceptic, regarded such a recommendation as potentially disastrous,
but happily impossible to follow. His scepticism consisted primarily
in a drastic re-evaluation of the powers of reason to underpin belief.
In the twentieth century the sceptic has mostly been thought of
as someone who denies that we know anything much; how close
to Hume he may be depends on Hume's concept of reason and
its relation to this sceptic's concept of knowledge.
In offering to show how the existence of a debate about scepticism
is a product of the concept of knowledge, I do not of course have
anything like this range of material in mind. That all the positions
commonly called scepticism form an intellectual natural kind, about
which one can aspire to generalise without falling into triviality,
is unlikely almost beyond consideration. What I shall try to do
is to account for the role played in this debate by a certain type
of argument, familiar to every student of philosophy from the first
of Descartes's Meditations, where it is exemplified by the thought
of the deceitful demon. The strategy is to invent a hypothesis with
two properties: first, it would make no noticeable difference for
us if it were true; secondly, if it were true virtually all our present
beliefs would be false. Then it is urged that, because of the second
property, we must either show that the hypothesis is not true or
abandon virtually all claims to knowledge; and that because of the
first property we cannot show that it is not true. Nowadays the
demon has a colleague in the scientist -whose computer feeds perfectly
'normal' stimulation to an excised brain floating in nutrient fluid.
Technology has advanced but the problem remains the same: how

to exclude the possibility that I am such a brain, how to claim any

knowledge of my environment if I cannot?
This style of argument fascinates, which is why we find so many
attempts to reply to it. But at the same time there is a feeling of
unreality about it, which is why so many of the attempts begin
in a posture of condescension, with the assumption that it must
be mistaken. Some, like Moore's famous response, 1 very nearly end
there. But still the feeling persists that there is a problem. Nozick
presented it as a major advantage of his analysis of knowledge that
it disposed of just this type of sceptical argument; in this he was
quite mistaken, 2 but the point remains that in 1981 he regarded
the job as still eminently worth doing. What I hope to explain is
the existence of both of these opposed reactions, and the apparent
difficulty of bringing about a decision between them. Up to now
I have been trying to account for features philosophers have dis-
cerned in the content of the everyday concept of knowledge; now
I consider something that characteristically happens when philoso-
phers talk about its range of application.
That way of putting the matter presupposes that there is some
connection here, that the everyday conceptual practice is at least
to some extent influential in shaping the philosophical debate. One
hypothesis, which by implication I have put forward elsewhere,3
suggests a different sort of explanation. A philosopher may work
under the influence of a picture of the nature of reality and of man
which assigns special status to a particular kind of relationship
between the human mind and the objects of its thought. And this
may incline him to reserve the honorific word 'knowledge' for that
kind of relationship, and so to deny that most of what ordinarily
passes for knowledge 'really' is such. In recommending his ideal
to us, he recommends a change of usage. (Conversely, his change
of usage is not arbitrary: it expresses his ideal.) And it is perfectly
possible that even his opponents may accept his usage, his way of
thinking of knowledge, and do so precisely for the purpose of show-
ing his ideal to be beyond realisation. If they do, they will end
up saying that there is no knowledge, or very little. What they
thereby reject, however, is not the everyday practice involving
'know' and its cognates, but the cognitive ideal associated with this

1 2
See G. E. Moore. See E. J. Craig, (3).
See E. J. Craig, (2), csp. chs. 1 and 2.
106 S E C T . XII

particular Weltbild, perhaps thereby the whole Weltbild as well.

It is not clear that che everyday practice is playing a formative part
in this debate at all; perhaps it is doing no more than lending it
a word.
One reply here, as usual, would be 'the proof of the pudding':
does it turn out, on inspection, that the same hypothesis that
accounts for so many of the facts about the everyday practice also
serves us well in understanding the status of the traditional sceptical
fantasies? But even before we start tasting the pudding there is reason
to think that not everything can depend in this way on the back-
ground metaphysic.
Someone who thinks that the background metaphysic is all there
is to it will surely be at a loss to explain the intuitive impact of
the Cartesian strategy. On this view the strategy is simply a covert
way of urging us to adopt different standards, those enthroned by
the ideal, and then show that these standards are hardly ever met;
it is not an attempt to convince us that hardly anything reaches
the standards we now set, therefore it poses no challenge to anything
we presently believe. If there were something else behind it, namely
the idea that the 'new' standards really are the ones we now set,
if only we would examine them carefully, or that they emerge from
the ones we now set as an obviously natural extension; if something
of this kind were in the offing, then that would be a different matter,
and might well induce the feeling that we were up against a sceptical
threat. That however would call for a rather different explanation,
one giving an account not just of the motivation for the philosopher's
standards, but also of the features of the everyday concept of know-
ledge (or whatever) which linked it to them.
What this means is that we have to look at another type of hypothe-
sis, one which finds seeds of scepticism in the everyday concept
itself, prior to any philosophical modification. It is not, as I have
in effect just said, at odds with a hypothesis of the former kind,
but complementary to it. For if a philosopher bends the concept
of knowledge to make it fit for duty at a particular station in his
system, one fact we should not forget is that it was that concept
which he elected to bend, rather than another, or rather than delimit-
ing a new concept and inventing a word for it; and there may be
something in the nature of the everyday concept that made it a
specially suitable choice. There may be, in other words, some feature
of the everyday concept which generates pressure towards

scepticism, or opportunities for it, and there might be something

in our philosopher's metaphysic or Weltbild which inclined him
to emphasise that feature. These remarks are obscure, partly because
anticipatory, but I hope they sufficiently indicate that we need not
think of the two styles of explanation as mutually exclusive.
Before trying to explain anything we had better pause for a longer
look at the facts to be explained; when we do it appears that what
we are looking for must involve (at least) two complementary com-
ponents. For the reasons given above it can hardly be the case that
the explanation lies wholly insulated from everyday practice,
depending solely on the properties of various thought-structures
in which only a tiny fragment of humanity have ever taken any
interest. That would make it look too much like an accident that
scepticism tends to be expressed in terms of the words 'know' and
'knowledge'. Besides that, there is the fact that when it is proposed
to usmeaning any of us who are prepared to think about such
thingsthat we hardly know anything unless we can rule out the
possibility of the demon, our reaction is not the same as it would
be if we were told that there aren't any doctors unless there is
someone who can cure any disease within two minutes.4 Most
of us don't much like the former point, but we find it all a bit
puzzling and, partly for that reason, irritating; the latter is just mani-
festly fatuous and arbitrary. Evidently the sceptical argument con-
nects m some way with something or other from the conceptual
practice that we all naturally grow up into.
On the other hand, it would surely be misplaced to want the
explanation of the existence of that conceptual practice to lead
straight into an explanation of the impact of the Cartesian argument
without any additional input. For in that event the question whether
anyone knows anything much, and divergence about the way to
answer it, would be a part of the everyday scenewhich it isn't.
So to account for its presence in philosophy we need either a further
factor not a part of every community's mental life, or another feature
of the everyday which inhibits it. And in the second case we should
still want to know what it may be in some philosophers that conquers
the inhibition. What is wanted, given the explanatory tools I am
employing, is an account of the likely genesis of features of the
everyday practice which offer handholds both to persons motivated
I borrow the example, though not the use of it, from Paul Edwards.
108 S E C T . XII

towards scepticism and to persons motivated away from it. Prefer-

ably, they should not be such as to award victory to either; for
if they were, one or the other would long ago have won.
A last preliminary point: some may think that the restriction to
the argument from demons and brains in vats robs my enterprise
of most of its promised significance. There are so many other routes,
it may be said, to radical scepticism. My reply, for which I shall
offer some reasons in Section XIII, would be: perhaps, but not as
many as has been suggested. If we understood the attraction and
repulsion of the sceptical strategy of Descartes's first Meditation,
we would have gone a long way.
In Section X we looked at the phenomenon of objectivisation
and the way it will act to transform the inquirer's original, subject-
ively relativised notion of a good informant. Of the four factors
we considered, numbers (1), (2), and (4) do not, so far as I can
see, have anything to do with scepticism or any tendency to scepti-
cism latent in the concept of knowledge. But what may well have
something to do with scepticism is condition (3): that the informant
should be as likely to be right about p as the subject's present con-
cerns require.
What we saw was that the objectivisation of this notion will tend
to push up the standards demanded of the informant's reliability.5
When we collect information, seek informants, or recommend
informants to others, we invariably do so in a state of extensive
ignorance; amongst the things we are very often, if not absolutely
always, ignorant about are the purposes for which the information
will be used, by whom, and under what conditions. That is to say,
we are ignorant of just those factors which determine the acceptable
levels of reliability in the wholly subjective case. We therefore adopt
a strategy designed to succeed whateverthe same response as orig-
inally introduced counterfactuality into the concept of knowledge
in this case: set a standard high enough to face the worst with.
Then we have vocabulary, and a style of speech (remember that
someone can in effect claim knowledge of something without men-
tioning the word 'know', merely by the way they announce it) which
marks this standard, and more cautious forms for offering infor-
mation of lesser reliability. The possibility of the worst compels
observance of the highest standard, and this is what the concept
See also Section XI.

of knowledge certifies.
As the reader will have recognised, I am now in danger of making
things much too easy for myself, for what these superlatives ('worst'
and 'highest') mean is still unclear, and they may mislead. The danger
is of slipping into saying that for knowledge, given that the concept
is governed by objectivisation in the way just envisaged, 'absolute'
objectivisation is called for, meaning by that the demand that the
informant be certain to have the right answer in any conceivable
world, including those 'sceptical worlds' of demons and brains in
vats; that the 'highest' standard, in other words, means a standard
capable of withstanding any conceivable mishap. That would explain
how and why the concept of knowledge brings scepticism in its
train; but of course it is not within a mile of being warranted by
anything we have established so far. It is rather as if we had said,
with Peter Unger, 6 that knowing (given its genesis) will come out
as 'being right as to whether p, and not by accident', and then
assumed permission to say that someone who is right, but would
be wrong if he were a brain in a vat, or a victim of the demon,
is only right to some degree by accidentleaving ourselves just
as open to the charge of misusing the notion of an accident as the
sceptic is to misusing that of knowledge. Perhaps he doesn't misuse
it, and perhaps this would not be a misuse of the concept of the
accidentalbut on that issue this cheap manoeuvre does not advance
understanding one inch. The hypothesis was that knowledge 'lies
at the end of the road of objectivisation', but where does that road
come to an end, or when have we gone far enough? The whole
issue turns on finding a justified answer to that question.
It should be said straight away that Unger's spectacular and notori-
ous argument for scepticism7 did not involve the cheap and ques-
tion-begging manoeuvre, but rather made use of a far subtler thought.
He began by pointing out a category of what he called 'absolute
terms'. Being flat, he said, means being perfectly flatnot at all
bumpya standard which, for all we know, quite likely nothing
whatever achieves, but which in all practical contexts is relaxed in
accordance with the degree of flatness needed for whatever purposes
are under considerationa rugby pitch, a lawn, a billiard table.
A necessary condition of knowing, he then argued, is being certain.

' See P. Unger, (1). See P. Unger, (2) and (3).
110 S E C T . XII

And 'certain' is an absolute term: being certain is being absolutely

certain, as being flat is being absolutely flat, and being a vacuum
is being absolutely empty. So we should consider 'know' in the
same light: we may and do relax the standard, but the standard
calls for the truth of the belief to be (absolutely) certain, so it is
truly satisfied only by a belief whose truth would survive even the
worst efforts of the demon. At most hardly any surfaces are flat,
and hardly any of the regions we declare to be vacua actually are
such; so in respect of both these terms we say a lot of things which
are false, but they are usually close enough to being true for their
falsity to have no adverse practical effects, given the purposes we
have in mind. Our claims to knowledge, Unger holds, are in the
same position. The sceptic is therefore right, though the point that
he typically insists on, that we do not know anything, makes no
significant practical difference.
It may look as if Unger's view would explain both the existence
of scepticism (which results from the perception of its truth) and
the resistance to it (which is a mistaken reaction to its practical
insignificance). His view is equivalent to the thought, expressed
in our terminology, that the objectivisation which forms the concept
of knowledge goes to the very limit of theoretical possibility. Every
conceivable way in which the belief that p could be wrongand
in the hardline version perhaps also all the inconceivable wayshas
to be taken into consideration. Since Unger's arguments for ascribing
such a sense to 'know', though far from negligible, do not wholly
convince me (nor, as we shall see, do they nowadays wholly convince
Unger), I would be more at ease with their conclusion if I could
see some practical motive which might push the process of objectivi-
sation to the theoretical limit in the way it implies.
Finding such a motive is not going to be easy. There is nothing
new in the point that human life takes place against a background
of beliefs, assumptions, presuppositions, or whatever we wish to
call them. That we are brains in vats, or victims of a Cartesian demon,
or that the deity is about to revise all the laws of nature, these
are thoughts we can just about entertainwith a bit of practice.
But it took a highly unusual thinker, conducting a highly unusual
investigation, to suggest them to us in the first place, and when
they are suggested our mind in a sense rules them out at once.
Nor is it surprising that nature should operate some kind of automa-
tic defences against thoughts which, taken seriously, could only

paralyse our powers of decision and action. That isn't yet to say,
as some opponents of scepticism have said, that they belong to the
pathological side of our mental life; but it is to say that they have
a role in it only under very special circumstances, which certainly
don't include the ordinary social practice of gathering and passing
on information, or certifying informants. In deciding whether Fred
can be advanced as an acceptable informant as to whether/) nobody
considers whether he would answer correctly were he (or they) sub-
ject to some systematic Cartesian illusion and then, having decided
that he would not, tries to calculate how much damage the fact
does to his credentials by asking how likely it is that that 'possibility'
will be realised. Just as importantly, nobody, however vital to them
it is that they get at the truth about/;, expects anyone else to consider
such matters before recommending an informant; nor does anybody
consider them before relying on one of their own beliefs. Unger
himself makes a remark which only sharpens the difficulty:
. . . with respect to the matter of whether there are elephants, for practical
purposes there is no important difference between whether you know that
there are elephants or whether you are in that position with respect to
the matter that you actually are in. 8
Why do we have the absolute concept of knowledge, must we not
ask, rather than the concept of this other state (the one we 'actually
are in'), or some vaguer concept taking in both, if 'for practical
reasons there is no important difference' between the circumstances
in which they apply? Or how do we arrive at it if the thoughts
it embraces (such as those of the classical sceptical possibilities) have
no role whatever in everyday mental life? The reply might be made
that even if we cannot answer those questions we can still point
to various 'absolute' concepts which uncontroversially have been
formedso there is no insuperable difficulty about their formation,
even if we do not see how it can happen. But this takes us only
half way, for in the case of knowledge there is a problem which
is not nearly so acute when we are thinking of emptiness or flatness.
I do not want to insist that 'flat' and 'empty' definitely are absolute
terms in Unger's sense. But if they are, as is on first impression
quite plausible, at least one can think of (reasonably) everyday pro-
cesses of thought which might have made them that way. Notice
that whatever processes our explanation calls upon do have to be
Ibid., (2), p. 201.
112 S E C T . XII

classifiable, without too painful a stretch, as 'everyday'; we are deal-

ing here with a thesis about the concepts of knowledge, flatness
and such, not with concepts that occur only in the work of certain
particular thinkers, presumably as a result of some thoughts or needs
peculiar to a very few. What is under investigation is equipment
supposed to be known and used by more or less everyone.
From this point of view, the immediately striking thing about
'flat' and 'empty' is that they have to do with series that are felt
to be seamlessly continuous. Once we have the idea of the contents
getting less, the bumps getting lower, the mind can hardly help
passing to the thoughts of even less, and lower still. And there is
no reason to stop where unaided perception gives out, nor would
there be even if it gave out at the same point for all of us. For
one thing, something which is perfectly flat to visual and tactual
inspection may turn out not to be flat enough for our purposes
when we try to use it as a snooker table, or as a mirror. And for
another, the images of things not available to unaided perception
which are made available by microscopy or a simple magnifying
glass are very like unmagnified images of familiar sized, ordinarily
bumpy objects. If we try to imagine what a surface which to us
appears flat is like for a very minute creature with perceptual acuity
suitable to the size of its body, we may very well use the same
imagery as we would use in imagining ourselves in a jagged, broken
The ways in which a belief can fail to be knowledge, however,
do not hang together in the same way. There is here no ordered
series of items or states of the same kind being successively revealed
by continuous improvement of some kind of intellectual sensitivity.
One could indeed speak in these terms of the passage from everyday
mistakes to the wholesale error induced by the demon, but it would
be a metaphor comprehensible only to someone who already knew
the literature of scepticism. The imagination is not lead smoothly
on in the same way from case to case. Without special circumstances,
or special prompting, I see no reason why it should ever make the
There is a problem, therefore, as to how the concept of knowledge
could have arisen, given Unger's account of it. But there is another
and sharper problem: why should scepticism have met with such
resistance? Consider the term 'vacuum', to which 'knowledge' is
alleged to be analogous. What would be the likely response to the

suggestion that scarcely any of the things we refer to as 'a vacuum'

actually is such? Surely the response, after a little havering perhaps,
would be that it is quite true: 'vacuum' picks out as it were an
'ideal' state to which the best of our vacua are good approximations
and that what approximations count as good depends on which
purposes the so-called vacuum is to serve. (And one might well
add that although the suggestion is quite true, there are not many
situations in which one would be doing anything more than making
a nuisance of oneself by insisting on it.) But a debate, with each
of two or more parties warmly advocating their own view, one would
not expect; and it is not in fact found. 'Are there nearly as many
vacua as we commonly think?' isn't an issue, and this is reflected
in the fact that many would prefer to reformulate the question: 'are
there nearly as many vacua as we commonly say there are?'with
the implication that we are perfectly well aware that we don't really
think what, if our words are taken quite literally, we often say.
If 'knowledge' were analogous to 'vacuum' the debate about scepti-
cism would be very different; most likely it wouldn't exist at all
people would quickly take the sceptic's point, and pass on. Not
only can one not see how an absolute concept of knowledge could
have arisen; there is positive reason to think that it hasn't.
Unger saw this problem, or one very close to it. He observed 9 that
when pressed to be accurate in our use of such terms as
'flat' we are inclined to say that we never did believe that (many
of) the various surfaces which we called 'flat' really were flat, only
that that they were flat enough for our then current purposes; but
that in the case of 'know' we feel, even if we are convinced and
finish up complete sceptics, that previously we really did believe
that we knew (most of) the things we verbally claimed to know.
But having pointed it out, Unger honestly admitted that he could
give no explanation, and that to give one 'must remain a further
question'. The answer, I believe, lies in giving up the thesis that
'know' is an absolute term, and then we shall have to take a more
complex, and less partisan, view of the phenomenon of scepticism
which is what Unger himself later came to do. 10
It seems then that there is no particular reason to continue with
' Ibid., (3), p. 89.
It will be seen that my discussion of Unger throughout this section relates
to his work up to 1976, when Ignorance: A Case for Scepticism was published.
Since then his views have changed a good dealsee Appendix, below.

the unpromising search for practical factors that might decisively

push objectivisation of the concept of knowledge to its theoretical
limitthe indications are that it has not, at any rate not unequivo-
cally, gone there. So we should pass to another highly pertinent
question: are there any such factors that would cause objectivisation
to stop at any definite and specifiable point?
We have seen the practical point of having a concept which requires
that a belief be subjected to fairly stiff tests; the concept is applied
only when the likelihood of its being false is very low. How low,
or what possible ways of being false are to be considered, is variable.
That the chief suspect might have an ulterior motive for confessing
to the murder will often be taken very seriously; that the booking
clerk might have an ulterior motive for telling me that the next
tram is due to leave m 5 minutes doesn't get a look in. (And this
is not because false confessions to murder are frequent and must
be allowed a fairly high probability, but because of what turns on
the outcome.)
Only people who antecedently expect a concept to be loaded with
all, strictly all, the conditions needed to determine its applicability
to each instance, will suppose that we could in principle find, in
the concept of the good informant (or of knowledge, for that matter),
what distinguishes the Old Bailey from British Rail. Far more attract-
ive is the supposition that the application of the concept is left to
judgement, and that in human adults of a certain level of intelligence
and experience judgement produces a workable degree of uniformity.
It is a matter of taking appropriate care, and no mechanical directions
for appropriateness are to be looked for in the concept itself.
Two thoughts occur. The one most immediately suggested by the
foregoing is this: that the 'practically explicated' concept will simply
be open-ended at this point. The concept of knowledge will make
room for any, even the Cartesian, degree of caution in its application,
provided there are purposes and circumstances to which that degree
of caution is appropriate. The second thought pulls against this one,
however. We have already agreed that the relevant level of caution
will be fairly highthe practical requirement of objectivisation will
put a fairly stringent lower bound on it. May there not be practical
requirements that put some sort of upper bound on it as well? If
so, do they perhaps shield the concept from confrontation with the
possibilities traditionally used for generating scepticism? In that
event, those who insist on confronting knowledge-claims with the

standard sceptical fantasies really would be straightforwardly mis-

using the conceptas has often been said of them.
Concentrating for the moment on the first of those questions,
we may say at once that there may well be some such requirements.
For a start, we have already observed that there are many 'possibili-
ties', including the now standard sceptical scenarios, which are just
not treated as such in ordinary (meaning: non-philosophical) prac-
tice. Not that they are considered and jettisoned; they are simply
not considered, the very thought of them is something strange and
new to any adult who comes across them, which the great majority
never do. Now if standard practice thus excludes them, will they
not also be excluded from any role in determining the applicability
of the standard concept of which, presumably, standard practice
is constitutive?
Further, can it not be argued that this is more than just a descriptive
point about everyday practice? Such extreme possibilities are not
worth considering for the powerfully practical reason that their reali-
sation would make no noticeable difference to us. Maybe virtually
all my beliefs would be false were the demon, or Berkeley's God,
in action. But then if the demon, or Berkeley's God, is in action,
acting on these beliefs will have, so far as I am concerned, just the
effect I wanted. There can be no practical advantage in trying to
distinguish between two situations which will be, so far as we can
tell, indistinguishable. And few enterprises are in this respect neutral;
where there is no advantage to be gained, there is nearly always
disadvantage in pursuing them. Had we unlimited time and energy,
then perhaps; but since we don't, their pursuit mostly distracts us
from what is more urgent. What we are looking at, after all, are
the factors giving rise to a very widespread feature of language
certainly not an invention of the leisured classes.
There may be some truth in that thought when it is applied to
certain sceptical hypotheses, such as for instance that which postu-
lates a demon absolutely consistent in his deceptions now and hence-
forth. But the more disturbing hypotheses, practically speaking, are
ones that propose some sudden change, a demon whose powers
(or intentions) are wearing out, a deity bored with the present order,
or grown jealous of the enormous respect accorded to Fred as an
informant. Here we are threatened by practical consequences
enough, but still such possibilities play no real part in our thinking,
no matter how important it may be to us to acquire true beliefs.

For an explanation of this, however, we can look in another direc-

tion: no practical purpose can be served by a test which every inform-
ant is bound to fail. We do have to find someone who satisfies
the test, for otherwise no informant, no belief, no action, no success.
And no success, no survival. Let it be granted that there are compel-
ling practical reasons for wishing to avoid falsehood, especially if
it might fall at an important junction in our whole system of beliefs
from which (like Descartes's rotten apple) it might infect many of
the others. But if action needs belief, having false beliefs is no worse
than not having any beliefs at all, so no practical motives could
lead us to prefer the latter state to the former and impose tests that
would expunge all belief before admitting a single falsehood. On
the contrary, what that would indicate is the rejection of the tests.
It looks as if the technique of practical explication will lead us to
a concept immunised by its own limiting conditions against the scep-
tic's threat; and conversely, if the concept of knowledge is vulnerable
to scepticism, no practical explication will be able to explain how
it got that way.
This, however, must surely be the wrong way to argue. We have
said that in everyday practice the idea of applying such tests as the
standard sceptical fantasies just doesn't come into anyone's head.
It follows that the opportunity to lay down the practice of rejecting
this kind of test, either for the reason of its unsatisfiability or any
other, equally just does not occur. We may conjecture about what
would happen to an attempt to introduce such tests into everyday
thinkingit isn't very hardbut that is not to say that everyday
practice is already determinate on the point. It has after all neither
need nor reason to be determinate. A decision isn't called for; no-one
raises the question, no-one even entertains it, let alone seriously
considers allowing such a test to affect the formation of his beliefs.
Societies have laws; but no society has a law against doing something
which it never enters anyone's head to do. (Thoughand here lies
the analogywere someone to think of it and do it, there could
well be a widespread, uncomfortable, and unenforceable feeling that
somehow or other there had been a breach of something.)
We must not be mislead. There are, in a sense, factors which
cause objectification to stop short of the sceptical fantasies: it is
bound to stop short of considerations which are never considered.
But that does not mean that the concept so formed acquires as it
were a 'hard boundary' at this point. On the contrary, the operative

fact is precisely that nothing happens here, so we neither have a

positive boundary nor the positive absence of one. The resultant
area of indeterminacy hosts the controversy about scepticism. On
the one hand, the genesis of the concept through the process of
objectivisation pushes us on towards acceptance of ever severer
tests and so finally over the edge and into the arms of the sceptic;
on the other, we are held back by the perception that the edge
marks the end of any contact with the practical requirements which
are the concept's ancestry. In much the same way, in a related area,
the impetus of objectivisation pushes us towards the thought of a
state of affairs which is in principle beyond our powers to recognise;
and pragmatism causes us a bad conscience if we give way.
None of this fully explains the existence of scepticism. Nor ought
it to, for scepticism is a philosophical phenomenon. Neither it, nor
its rebuttal, forms any part of everyday practice, where the whole
issue is simply unknown. And my style of explanation is suited
only to explaining the state of everyday practice. What, drawing
on the technique of practical explication, I have tried to make plainer,
is the nature of the gap in the everyday concept in which the debate
about scepticism can go on, should any philosopher find motive
to start it up. The motive might be the prevalence of some metaphysic
which suggested a particularly insightful grasp of reality. But it might
equally be reflection on the everyday practice itself. For we do find
there a tendency to seek to improve the reliability of our information,
so that if the proponent of the sceptical argument says that he is
merely inviting us to do a little more of what we normally do anyway,
no decisive refutation of him will be available. He can extrapolate
this tendency without doing anything which the concept of itself
forbids; and such extrapolation will lead to scepticism. His oppo-
nents, on the other hand, can find in the normal rationale of everyday
practice a reason for crying halt to the extrapolation before an irreme-
diable scepticism sets in; their perception that some kind of non-
arbitrary boundary has been crossed is no illusion. No more than
the radical sceptic do they do anything which the ordinary concept
of knowledge disallows. Quite the contrary, each of them does some-
thing which certain (and of course quite different) aspects of the
concept encourage.
To try to go further and treat of the specific motivations which
have lead some philosophers to exploit this situation would
pass beyond the scope of this essay; it would mean introducing
118 S E C T . XII

considerations which cannot possibly be thought to be part of the

mental life of any and every human community. But since the notion
of practice has been much to the fore in our argument, I shall end
the section with a brief excursion designed to test the assumption
that whatever the motives -which might lead a philosopher to scepti-
cism, they could not be ones having any practical bearings.
A plausible view, which I have already alluded to, is this: there
are indeed very compelling practical motives for trying to avoid
falsehood, especially if the falsehood might have the kind of centrality
that could make it likely to infect many of our other beliefs. But
survival calls for action, and action needs belief, so having false beliefs
is no worse than not having any beliefs at all, and will often turn
out a great deal better, hence no practical motives could lead us
to prefer the latter state to the former and impose tests that would
keep out falsehoods by rejecting everything.
On reflection, however, that argument can be seen to distort the
order of investigation. If we are to be true to the notion of a practical
motive we must not forget that practical motives have to operate
in actual circumstances, and in actual circumstances we don't know
in advance of applying a test what results it is going to giveor
we would not be applying it. The Cartesian procedure opens with
the thoroughly practical intention of avoiding falsehood. To make
as certain as possible the attainment of this very practical aim it
subjects beliefs to the strictest test it can devise. That is practically
minded too: if we are going to do it let us have no half measures.
If it now transpires (contrary to Descartes's own opinion, but in
accordance with the views of many others then and since) that the
test is too strict, and that hardly anything survives or proves recon-
structable, that does nothing to show that the test was not in the
first place practically motivated but at most that there can be no
practical motivation for allowing it to dictate one's beliefs, and
especially one's withholdings of belief, now that the outcome is
known. Would the practically minded man have applied the test?
Quite likely he would, unless prescient as well as practical.
But another aspect must not be forgotten. Perhaps he didn't need
prescience after all. Perhaps something other than foreknowledge
of the result of the test could have told him that there was something
wrong with his enterprise, if conceived in the practical spirit of an
attempt to eradicate falsehoods from the belief-corpus. Perhaps he
could have seen that there were certain beliefs which he was not

going to drop, no matter what the test told him about their status.
If that is so, the position is different. It isn't exactly that there is
no practical motivation for the test, but rather that the motivation
is balanced by the inability to take advantage of it. There may be
excellent reason for me to go to the shopsI need to buy something.
But then there is the countervailing reason that I am too mean to
buy anything when I get there. So I may just as well not go; and
likewise if there are certain beliefs which I am not going to change,
I may just as well not bother to investigate them.
The to-and-fro of the debate doesn't finish there, however. Other
possibilities remain uninspected. For one, perhaps he couldn't have
seen, without applying the test, that certain beliefs were going to
stay with him no matter what its verdict. For another, perhaps in
order to test those beliefs which he was capable of changing, he
had to operate test procedures radical enough to induce doubt, or
in some cases pseudo-doubt, in respect of all his beliefs, immovable
ones included. For a third, even in the case of such 'immovable'
beliefs, there may be practical benefits arising out of our attitude
towards them and other beliefs consequential on them if they are
subjected to a very rigorous test and are seen to pass it; in that
event there could be a practical point in applying the test to them,
even at the risk that they would fail it.
We are ill-equipped to deal with these thoughts, partly because
we have too ill-defined a notion of what constitutes a practical ben-
efit, and partly because (as in the second and third possibilities just
mentioned) we are not clear enough about the boundary-conditions
of the whole inquiry. But appeal to our ignorance, of which there
is plenty in this area, will only serve to stave off an attempt to
show that no practical motives could lead to the formation of a
concept of knowledge capable of supporting, or in some way permit-
ting, radical scepticism based on the argument from the thought
of the demon. It cannot help with the positive task of showing how
practical pressures could give rise to such a concept. That task is
evidently beset by all manner of difficulty; there is no obstacle here
to understanding the feeling that the argument from the possibility
of the Cartesian demon invites us to abandon the very purpose of
speaking of knowledge.

Two ideas about the way in which philosophical scepticism arises

enjoy at present considerable currency. One often hears of the 'first-
person' approach to epistemology; its locus classicus is held to be
Descartes's first Meditation: the adoption of the stance of unaccom-
panied epistemological soloist is followed by catastrophic sceptical
collapse. True, with one bound our hero is free, but that part of
the story fools nobody nowadays, and didn't fool everybody at
the time. The first of the two ideas, then, is that the stance is what
causes the collapse. The second, associated in particular with Bernard
Williams and Thomas Nagel,' is that scepticism is the offspring
of an 'absolute' conception of truth, which encourages the attempt
to know reality as it 'really' is independently of the perspective,
capacities or interests of any observer. It is the search for what Nagel
more figuratively calls 'the view from nowhere' that makes sceptics
of us. I shall argue, on the contrary, that neither of these ideas
is sufficient to explain the pull towards scepticism, nor the repulsion
from it. I do not say that they have no part to play, because I
do not think that there is just one route by which scepticism is
reachednor indeed just one position of which 'scepticism' is the
univocal name. What I do say is that they require, at least, supple-
mentation of the kind I have suggested in Section XII: something
or other must be offered to explain why the standard sceptical stones
(about demons, brains in vats and so on) should be thought to have
anything to do with the business of knowledge, and why their con-
nection with it should at the same time be felt to be so questionable.
The first-person approach is sometimes characterised as that which
takes as its central question 'What do I know?' rather than the imper-
sonal 'What is known?'. But that characterisation doesn't by itself
capture any important distinction. In answering the former question
I might be prepared to make uncritical use of 'what is known',
the contents of almanacs and encyclopedias; in answering the latter
Esp. B. A. O. Williams, (2), ch. 1, and T. Nagel, ch. 5.
SECT. X I I I 121

on the other hand I might start by taking 'what is known' to be

the aggregate of everything of which some individual can rightly
say 'I know it'and then all would turn on what I took to be
the proper way of tackling the first-personal inquiry. It is a particular
way of tackling that inquiry, I believe, which philosophers have
in mind when they speak of the first-person approach to epistemo-
logy. In deciding what I know (and how I know it) I am not to
make use of any material except what I can myself stand as authority
for; no other person is for me a source of legitimate belief unless
I am in a position to certify him as such.
A fairly platitudinous observation is that this restriction is likely
to shorten the answer to our question 'What do I know?' a good
deal. But that has little to do with scepticism. For scepticism to
come into view we need the answer reduced to 'Very nearly nothing',
or 'Nothing about the physical world', or something else of like
brevity. If the first-person approach can do that it must have muscles
we have not yet seen.
Here an aside: an argument has recently come into fashion, found
by Kripke in Wittgenstein, which purports to show that an isolated
individual cannot operate a language, but will be overtaken by what
one might call semantic collapse. That would be quite catastrophic
enough to count as a severely sceptical result, maybe the ultimate
in scepticism. But whether it really follows, and on exactly which
reading of 'isolated', are questions which mercifully need not detain
us now; the history of the topic sufficiently instructs us that it isn't
from arguments about rule-following that scepticism has emerged.
The view that the isolated individual will never know anything
because he will never get any concepts working, whatever its claim
on us, is not what we are looking for at the moment.
Epistemologists are understandably afraid of the principle (which
I shall continue to call the Iteration Principle as in Section VIII)
that one only knows that p if one knows that one knows that p.
They fear regress, both of beliefs and of reasons, which takes us
far beyond what is actually the case and probably also beyond what
possibly could be the case; down that hole lies radical scepticism.
One may stop the regress by digging supposedly self-intimating
foundations, but the only foundations with any claim to being self-
intimating have turned out too flimsy to support much of a build-
ingwhich brings us back to scepticism again. I agree, but I would
point out that any connection between the acceptance of this princi-
122 SECT. X I I I

pie and the adoption of the first-personal stance is very dubious,

not to say wholly imaginary.
Does the first-person approach lead to the Iteration Principle and
all concomitant difficulties? It could be thought toand evidently
has been, as we saw in Section VIII. If I ask 'Do I know whether
p', and want to settle the matter myself, I will stop only when I
have assured myself that my state (in most cases probably something
about how I acquired the belief) is one which correlates well with
being right on that sort of question. And that may look like saying
that I will stop only when I reach the state of knowing that I know
that p. Let us accept that; it surely does not amount to saying that
I know that p entails that I know that I know that p. If it did,
we would be faced with an absurdity. For consider the simpler case
in which I, adopting the Tm going to settle it myself posture,
ask 'whether/??' Again, I will stop only when I have assured myself
that p, and if I do it responsibly that will happen when I have got
myself into a state that correlates very well with being right on
that question, one which may fairly be called 'knowing that/?'. But
no-one will think (please!) on these or any other grounds that p
entails 'I know that/)'.
It may be said that I have to do a little more than that. If I responsi-
bly investigate the question whether/; then I must try to get myself
into such a state and be convinced that I have succeeded, so that
if, having carried out the investigation, I am asked whether I know
that/?, I will reply that I do. Let us agree; surely still nobody thinks
on those grounds that p entails that I (or that they) know that/??
The truth, altogether less barbed, is that anyone who has carried
out a successful investigation will then know that p (or not-/?, as
the case may be); and anyone who thinks that they have carried
out such an investigation will (understandably) think that they know
that/?. Hencenot as any consequence of the concept of knowledge
but merely as a special case of the aboveanyone who has carried
out a successful investigation into whether or not they know that
p will then (assuming that they do know it) know that they know
that/?. But nothing about this suggests for a moment that one cannot
know that/? without knowing that one knows. The Iteration Princi-
ple says that if one knows that p one knows that one knows that
p, which is exciting and dangerous, but unfounded. What we have
just reached says that if one knows that/7 and has successfully investi-
gated the question whether one knows that/7, then one knows that
SECT. X I I I 123

one knows that p; and this is very well founded, and completely
harmless, and really quite boring. Its content is no more than this:
that successful investigations result in knowledge.
There may however be a line from the first-person stance to scepti-
cism which does not pass through the Iteration Principle. Perhaps
it is not quite the form of scepticism we have had in mind in the
paragraphs immediately preceding, but it is still quite sceptical
enough to deserve the name. If we have adopted the first-person
restriction and are interested in whether we know that p then it
will not satisfy us just to know that p so long as the fact that we
know that p remains opaque to us. Suppose that someone is in
a position to assure mequite likely it will be on the basis of an
externalist analysis of knowledgethat provided I do have hands
(provided neither demon nor computer-aided scientist is at work)
I know that I have hands. Nowhe will typically continueyou
believe, indeed you have no shadow of doubt, that you do have
hands, so without casting any greater shadow of doubt you may
proceed to believe that you know it. I am offered a conditional
with the observation that I am incapable of doubting the antecedent;
and if I accept the conditional I had surely better abandon hope
of ever showing that I don't know that I have hands, since for that
I would have to show that I don't actually have any. But something
seems to have gone missing, for I knew when I began that I had
all these beliefs and could raise no real doubt about them; but I
was worried by the thought that I could be in that state and the
beliefs be false, for instance if I were subject to the demon, or were
the consciousness of a brain in a vat. I wanted to get behind the
fact that I have these beliefs and cannot doubt them, to see whether
I could find a guarantee of their solidity which indubitability alone
cannot provide, not, Moore-like, to re-affirm the beliefs and then,
via an externalist analysis of knowledge, to conclude that they were
known to be true as well as true simpliciter. And in that I failed:
faced with the hypothesis of the demon and its like, I found myself
just emptily repeating my own convictions without being able to
point to anything on which they were based which would not equally
be present were they false and any of the traditional sceptical hypoth-
eses true. Isn't this scepticism? And doesn't it emerge from the first-
person stance?
I shall not spend time on the first of those questions. To the
second I would say 'No'. No, because the first-person stance by
124 SECT. X I I I

itself does nothing to explain why the fact that I can give no dis-
tinguishing marks by which to rule out that sort of possibility should
have any significance for me at all. And once we allow that sort
of possibility a place in our thoughts we will get the same effect
whether we operate from the first-person stance or not. It can easily
come to look otherwise, for if we ask 'Does Fred know that/;?',
why should we not use our knowledge that p as a baseline and
start off by observing that p is true? None at all (that we have seen
so far), unless we are going to allow the sceptical hypotheses their
traditional role. Once we do that we shall find ourselves stuck on
the thought that our decision that Fred does or does not know
that/) rests on our decision about p, which in turn rests on a belief
which we hold in spite of the fact that, in the sceptic's story, it
would be false although to us the world was indistinguishable from
the world we are actually inas we at any rate immovably believe.
Another point arises here. Once we have decided to admit the
requirement that we be able to distinguish the circumstances involv-
ing the demon from those which, as we believe, actually obtain,
we do not then need the first-person approach as an extra: we get
it thrown in. For if we see the challenge in those terms there will
be no point in responding by asking about someone else's know-
ledge, let alone trying to take someone else's opinion as ground-level
evidence: even that there is anyone else will be a belief we can only
be entitled to make use of once the challenge has been met. That
Descartes first introduced himself in isolation and then thought of
the demon is from this point of view an expository accident; he
could just as well have thought of the demon and then have realised
that this was one he was going to have to cope with on his own.
How may we summarise? It isn't (obviously) that first-persona-
lism in epistemology doesn't exist, nor that it has nothing to do
with scepticism. The point is that both first-personahsm and scepti-
cism result from thinking about certain possibilities (fully compre-
hensive dreams, demons, brains in vats, in general any hypothesis
on which our cognitive faculties would give a systematically distorted
view of large tracts of reality) which are normally ignored. If we
want to understand scepticism we need to understand how and why
such possibilities come to be taken seriously.
The other suggestion I want to discuss in this section is the idea
that scepticism might be the outcome of the attempt to achieve an
'absolutist' view of truth. Bernard Williams writes of the 'absolute
SECT. X I I I 125

conception of reality', which he sees as a possible (concealed) source

of Descartes's scepticism; Thomas Nagel writes of the attempt to
reach an objective view, the 'view from nowhere', which shall have
left behind all beliefs that are due to our particular character and
situation. I begin by examining Nagel's position as found in chapter
V of his book of that title. Here is a suitable quotation to start
Objectivity and scepticism are closely related: both develop from the idea
that there is a real world in which we are contained, and that appearances
result from our interaction with the rest of it. We cannot accept those appear-
ances uncritically, but must try to understand what our own constitution
contributes to them. To do this we try to develop an idea of the world
with ourselves in it, an account of both ourselves and the world that includes
an explanation of why it initially appears to us as it does. But this idea,
since it is we who develop it, is likewise the product of interaction between
us and the world, though the interaction is more complicated and more
self-conscious than the original one. If the initial appearances cannot be
relied upon because they depend on our constitution in ways that we do
not fully understand, this more complex idea should be open to the same
doubts, for whatever we use to understand certain interactions between
ourselves and the world is not itself the object of that understanding. How-
ever often we may try to step outside of ourselves, something will have
to stay behind the lens, something in us will determine the resulting picture,
and this will give grounds for doubt that we are really getting any closer
to reality.2
There seem to be two ideas playing leading roles here. One isto
let Nagel's metaphor go on running for a bitthat Nowhere is not
a place on our map. In other words, whatever intellectual twists
and turns we make, the result has still been reached by us, with
our properties of mind and body. There is no way of reaching a
belief without some of these playing a part in the process. The other
idea is that of a theoretical advance which gives us good reason
to believe, in particular cases, that certain apparent properties of
reality are in fact contributions of ours, and should therefore have
no place in our new view of what the world is objectively like.
(A classic case, to which Nagel later refers, is that of secondary
qualities: we form a conception of the world and our perception
of it which explains why we perceive things as coloured without
saying that they really are coloured.) The second of these, so Nagel
Ibid., pp. 67-8.
126 SECT. X I I I

implies in the quoted paragraph, leads us to the first, and the first
(this seems to be said in the final sentence) gives us grounds for
scepticism. We can't reach Nowhere, so the possibility of being
wrong is ineradicable.
We need to be clear just how the train of thought is supposed
to run. Why, for instance, should we be so keen on that last sentence
(of my last paragraph)? Our beliefs are formed by usthat seems
to be the literal content of the metaphor about Nowhere. Now
if that thought, unaided by hidden premises, has sceptical conse-
quences, then any belief-forming being must be subject to scepticism.
The idea that God stands above scepticism, which afflicts only crea-
tures, will have to be dropped, and the doctrine of his infallibility
declared incoherent. One can avoid this conclusion only by identify-
ing the facts with God's beliefsso no wonder that something of
the kind appears to have tempted some theologians. The hidden
premise, that the belief that p is distinct from the fact that p, is
the further, minimally realistic, premise which we are to add to
'Our beliefs are formed by us'. Nagel, it appears, is happy to add
the slightly stronger premise that we form these beliefs in interaction
with the facts, but that strengthening can hardly be the decisive
factor in the production of scepticismwhat if we formed them
without interaction with the facts?
That premise is also part of the argument from the objectifying
theoretical advance. We are to envisage a sequence of events, thus:
(1) The World (W) appears to me to be Fand I accordingly
believe it to be F.
(2) I form a theory (T) about Wincluding mewhich explains
(amongst many other things) how W comes to appear F to
(3) According to T, W is not F.
(4) I come to believe Ton account of its great explanatory
powerand give up my belief that W is F. (Whether or not
it continues to appear to me to be F.)
Now it is at this stage that Nagel arrives (see the passage quoted
above) at the thought: 'But this idea [I take it, the corpus of beliefs
that constitute believing T], since it is we who develop it . . .' He
goes on: Tf the initial appearances cannot be relied upon because
they depend upon our constitution in ways that we do not fully
understand, this more complex idea should be open to the same
SECT. X I I I 127

doubts .. .' But that is not obvious; at least, it is not obviously

a legitimate conclusion to draw from the fact that we have just exper-
ienced the sequence (l)-(4). For nothing rules out the possibility
that we shall, going on in pursuit of our fully objective view, arrive
at a theory Tn, likewise about W (and ourselves within W), which
explains everything that happens in W, including everything that
happens in us. And 'everything that happens in W includes of course
the full story of how we started off with (1)and suchlike appear-
ancesand ended up with Tn. It goes without saying that it is we
with all the properties that Tn attributes to us, and indeed by
virtue of some of themwho have arrived where we have arrived.
And it is we who are now reviewing that factbut then our theory
explains how we can do so. And this theory, let us then suppose,
shows every sign of withstanding all subsequent discovery and exper-
No doubt all this is wildly fanciful. But I cannot see that any
contemplation of the route from (1) to (4) can show it to be impos-
sible. And just suppose it were to happen, what would then remain
of Nagel's metaphor that 'something will have to stay behind the
lens'? In so far as we have good reason (which ex hypothesi we
have) to believe Tn, we have good reason to believe that we have
finally grasped everything about ourselves, as well as what every-
thing else is like independently of us.
Scepticism shouldn't arise from the thought that this complete
and wholly objective theory could never be formed. Perhaps it could,
for all we have seen. It may arise from a thought quite different
from the problem of completing the investigation of which (l)-(4)
represents one stage. Perhaps the whole of this enterprise, however
successful it may at some future time appear to have been, is dogged
by an invisible dog. Though invisible, the dog is nevertheless very
familiar: one of the traditional sceptical possibilities. Maybe the best
supported, most complete theory we could even imagine getting
would still be radically misleading; maybe there is some hitch in
our processes of belief-acquisition for which we could never have
any evidence.
If this is what we are left with, things are rather different. If scepti-
cism depends on a concern for that kind of possibility, we still lack
an answer to the question why anyone should ever have been con-
cerned about it. It is not clear, for all Nagel's exertions, that it
emerges as an inevitable by-product of the drive towards the objec-
128 SECT. X I I I

tive view, if that is to be understood as the advance illustrated by

our sequence (l)-(4). It may nevertheless have something to do with
objectivity, when that is characterised slightly differently, and that
is what I have in effect suggested.3
The business of the objectivisation of a concept might in a (very
limited) sense be thought of as involving the formation of an objective
theory of the world. To have even a minimally objectivised concept
I must come to think of the world as containing other perspectives
than the one that I have here and now. For one thing, it involves
the recognition that there are other people, and that they may make
(and with luck also pass on) judgements about objects which are
available to me, constituents of my world; further, that these may
be judgements which I, from my position and circumstances, am
at present not able to make. If I am to benefit fully from their powers
of judgement, I shall need a little more sophistication still: their
judgements cannot always be taken over by me exactly as they made
them, but need to be adjusted to take account of certain differences
between their situation and mine. (Indexicals provide the commonest
example: we must learn to judge their reference with an eye to the
spatio-temporal location of the speaker.) In acquiring these capacities
we are building a theory (at this stage a very elementary one) about
a common world and the way in -which different views of it lead
to different judgements about it. This, it seems, is the same process
as the one that Williams and Nagel are interested in, though possibly
an earlier phase of it. 4
One may think of these varying perspectives purely cognitively,
as different vantage points from which different individuals or groups
survey the facts. But we should also think of them ideologically:
the different vantage points may, and frequently will, be associated
with different capacities and purposes. It is this idea, I believe, that
we need to engage if we are to get a view of those aspects of objectivi-
sation that may point in the direction of scepticism.
Williams imagines two subjects, A and B, each claiming know-
ledge of aspects of the world, but representing it differentlyperhaps
because of something so simple as a difference of position, perhaps
for more complex reasons like a difference of perceptual equipment
or conceptual resources. If we accept their claims as claims to know-
Principally in Section XII.
Possibly an earlier phase than concerns Nagel; Williams(2), p. 64appears
to have this kind of thing in mind when he writes of a Very primitive example".

ledge, he goes on, we need to 'form a conception of the world which

contains A and B and their representations'. This is the same starting
as that of Nagel that we examined earlier (although he expressed
his problem in terms of our point of view, Williams more imper-
sonally): we are to aim at a theory of reality which will explain
why it seems to be F from a particular standpoint. And his finishing
point is the same as well. Any picture we may come up with
... is open to the reflection, once more, that that is only one particular
representation of it, our own, and that we have no independent point of
leverage for raising this into the absolute representation of reality. 5

No more than Nagel's (to all intents identical) argument does this
lead to radical scepticism. What it describes is a particularly sophisti-
cated version of the familiar process of correcting our own mistakes.
It might sharpen our consciousness of the claims of fallibilism, and
make us more alive to just how many of our present beliefs might
have to go by the board before cognitive stability is reached. But
that it can be reached, and that our investigative practices are leading
us in the right direction even if the road may be longer and more
disturbing than we thought, this need not be brought into question.
Just why any view should be open to the reflection that it is 'only
one particular representation' still stands to be explained. It is not
explained by the undoubted fact that some views invite that reflec-
The idea that is here broachedthat of the attempt to get a view
of the world which will enable one to take account of the perspectives
of othersdoes however have something to do with the debate about
scepticism, and does have something to do with an important source
of encouragement for the sceptic. In this Nagel and Williams are,
I believe, right. But what does the trick is the way in which this
procedure of objectivisation, as I have called it, acts to form the
concept of knowledge; whereas they see it as it were pushing our
conception of the world out of reach, always one step beyond the
latest advance of our inquiry. What does not emerge (and cannot
emerge from that style of argument) is why they think it impossible
in principle for us ever to catch it up.
Williams (2), p. 65.

It has long been a complaint that some of our colleagues, especially

Great Dead Colleagues, tend to treat epistemology as if human
beings were spectators of reality rather than active constituents of
it. Such complaints have not often been very sharply focused, and
terminological oppositions like that of the standpoint of the Carte-
sian consciousness versus Bemg-in-the-World have done little to
help. We are now in a position to offer a small truth from the area
towards which these high phrases foggily gesture. It is a modest
truth, and hardly justifies the woolly charge against the Great Dead
Colleagues, but at least it is true, and can be made tolerably perspi-
cuous. Assuming agreement that the concept of knowledge calls
for truth of belief across a range of circumstances, some of which
at least must be non-factual, we can say that it bears, in this feature,
a mark of its origins m the needs of an active being. The right sense
of 'active' needs some care, however.
Why we should want to believe truths is a question we will come
back tofor the moment let us start from the fact that we do. The
decisive point is that we have a degree of choice as to what, and
whom, we believe, and can therefore adopt strategies intended to
improve our chances of coming to believe truths. Since these strate-
gies are to operate, as we have seen, under conditions of partial
ignorance, they need to be such as will work in a variety of con-
ditions, not all of which can turn out to be realised. Suppose however
that our beliefs were simply triggered in us by the environment,
and further that there was no question of our intentionally selecting
those circumstances most likely to trigger a true belief on a given
question. Then there would be nothing left for us to do but simply
to hope that the environment would treat us well and deal us true
beliefs where we needed them.
Under such conditions a concept with counterfactual implications
would be redundant. It is true that a partially ignorant being cannot
pointfully try to produce an effect without adopting a counterfac-

tually effective strategy; but one can perfectly well hope for an effect
without being in the slightest bit concerned whether or not it would
have occurred had circumstances been in any way different. Indeed,
hoping that it would occur under circumstances differing from the
actual sounds like anotherand rather strangeenterprise, perhaps:
hoping that its occurrence is probable. What the point of doing
that might be, as distinct from hoping that it would occur, isn't
at all clear: what might impel us to entertain a hope which would
be satisfied if the event proved to have been likely, even if it didn't
actually happen? I don't say that there could never, in any particular
instance, be an answer to that questionsometimes one might be
less concerned to be right than to avoid looking foolish. But that
is not an answer so natural and comprehensive as to make us expect
the counterfactual component of the concept to be as it is even if
no active truth-seeking strategies were involved.
Notice that this argument does not call upon the fact that, gener-
ally speaking, we want our beliefs for certain purposes, to enable
us to act successfully in certain ways. No doubt that is so, no doubt
it has some significance for the theory of knowledge, and a great
deal for the sociology of knowledge. But all that the present argument
makes use of is the fact that we are active, mentally and otherwise,
in seeking true beliefs; and it would hold even for a being whose
interest in true beliefs, once acquired, was purely theoretical.
That brings us back to asking: why should we be interested in
believing truths? This wider question has a clear place here, and
a fundamental one. Counterfactuality may be a central feature of
the concept of knowledge, but we arrived at it only as an inevitable
feature of any well conducted search for truth. So why are we search-
Imagine someone who watches certain events without the slightest
need or wish to intervene in themthey do not affect him in any
way that would cause him to prefer that the series should be just
as it is rather than otherwise, or vice-versa. His natural equipment,
let us suppose, causes him to form beliefs about what is happening.
Since, ex bypothesi, it is a matter of complete indifference to him
which events occur, and we have so far provided him with no other
evaluatory principles, there is as yet no reason in sight why he should
prefer these beliefs to be true rather than false, or indeed why he
should be pleased (or sorry) that he has beliefs at all. What do we
have to add to this neutral state for it to begin to make some difference
132 S E C T . XIV

to him whether his beliefs about what happens are true or not?
For a start, let us simply suppose that he prefers some events
to others. If he is right in believing that A is about to occur, and
he likes A, then something nice is coming up, if he is wrong then
it isn'tand that makes a difference to him. But so long as we take
it that he cannot actually affect what is going to happen, and don't
feebly abandon our inquiry by assuming that he values being right
per se, this still doesn't throw any light on why he should want
his beliefs to be true. If A happens, then something nice happens,
but whether he antecedently predicted A does not, in this very simple
model, make A any nicer, or the non-occurrence of A any nastier.
As soon as we get pleasures of anticipation and pains of disappoint-
ment into the model, the situation will of course change in this
respect, so this one might call the first grade of cognitive involve-
Given this grade of involvement, it does become the case that
if he can exercise any sort of control over what he believes, he can
affect something which it is in his interest to affect. Now he has
a motive for selecting, if he can, between different ways of acquiring
beliefs, but it is not yet by any means clear that he will choose
to select beliefs for their likelihood of truth. Having a true belief
that something nasty is inevitably on the way may well make life
a good deal nastier than the nasty thing will in due course make
it anyway; for that matter, having a false belief that something nasty
is coming will be no better, up to the moment when it is falsified.
He may decide that the pains of anticipation of the nasty are worse
than those of disappointment at the non-eventuation of the nice,
and opt so far as he can to believe in a thoroughly rosy future irrespec-
tive of its likelihood; or he may find it so enjoyable when things
turn out better than expected that he aims at a deeply pessimistic
set of beliefs in the hope of thus maximising the number of pleasant
surprises; or he might prefer to avoid believing anything as far as
possible. To explain the preference for truth from this standpoint
we would therefore need a supplementary, and I would have thought
far from compelling, hypothesis about the balance between these
various (and presumably variable) psychological forces.
A second grade of involvement comes if we suppose that he can
in some cases affect what happens. Then at last it does appear that
he must come to value true beliefs. For he needs true beliefs about
what will happen if he undertakes such and such a course of action,

as well as about what will happen if he does nothing, but just lets
things slide. Once this stage is reached he must develop an interest
in training himself to form his beliefs in ways which reliably lead
to truths. And if we also, as for any hint of realism we must, suppose
him to be partially ignorant of the conditions under which he is
operating, reliability must include the counterfactual property that
the methods would still have produced true beliefs had the world
been somewhat different.
Truth, and the counterfactual effectiveness of the processes used
to reach it, are two of the principal features of both the concept
of knowledge and that of a good informant. Active involvement
enters twice, on my account: we want true beliefs because we are
agents; and we actively seek the truth, which is why we must try
to 'track' it, not merely hope to hit it. I cannot decisively rule out
the possibility that these two features may be convincingly explained
without calling on the fact that the concept of knowledge is a concept
formed and operated by active beings who need to direct their
activity. And I grantat least I can find no argument to the con-
trarythat a pure spectator could in theory operate it. But why
should he? To ignore the question is to accept that there is a practice,
common to the vast majority of human beings, which mysteriously
exists without any basis in the human situation. Answering it, on
the other hand, won't be easyfor anyone who thinks that episte-
mology deals with a purely spectatorial side of our nature.

Certain writers have recently drawn attention to a principle about

the transmission of knowledge, namely: that if someone who knows
that/; tells me that/", I myself then know that/*. It codifies a wide-
spread intuition, or perhaps I should say a widely accepted linguistic
practice, according to which it is perfectly normal to speak of
someone as knowing something just because he was told it by
someone else who knew it. I don't acquire much, if any, understand-
ing of the facts just by being told them; but I do, according to
this practice, at least come to know what they are.
Now the obvious first impression is that the approach via the
good informant will account for this feature of the concept of know-
ledge, for if a good informant on the question whether p tells me
that p (or that not-p, as the case may be), do I not then myself
become a good informant on that question? It seems so. For I am
then as likely to be right as was whoever told me. If he was a good
informant therefore, so now am I. And a second obvious point beck-
ons: if he didn't know whether/? then neither do Iunless I know
itanyway, independently of his telling me. Likewise, if my informant
was not likely to be right then neither am Iunless, again, for some
independent reason I am likely to be right anyway. There certainly
appears to be a promising symmetry here. In this section I shall
investigate to what extent it will hold up under closer scrutiny.
We shall find a good many lessons learnt earlier coming round again.
Let Fred be a good informant as to whether p. He has some
property the possession of which makes it very likely that his opinion
on that question will be true, and indeed it is true. Now suppose
he tells Mabel. Is she now a good informant as to whether/??
If that is strictly all that happens, then of course she isn't. For
her to be a good informant it must be the case that she will now
tell us the truth, and that isn't yet guaranteed; Fred's assertion that
p (suppose for convenience of exposition that that is the truth in
question) may have had no effect on her whatsoever, and she may
SECT. XV 135

not otherwise have been disposed to tell us that p. Clearly we need

to add something or other to the situation to make sure that she
will have anything to say at all, let alone the truth. The absolute
minimum we can add is that she is indeed disposed to tell us that
pwhich is the same as Fred told her.
That still leaves it open, however, for her disposition to have
nothing to do with her interaction with Fred; perhaps she would
have told us that anyway. In that eventso at least runs the natural
reactionwhether or not she is a good informant as to whether
p has nothing to do with his having told her that p, hence nothing
to do with any questions about transmission. I am sure the reaction
is right, but it may be worth a couple of paragraphs to follow it
To be judged a good informant, Mabel will need to be disposed
to tell us the truth, and also to have an 'indicator-property'one
which correlates very well with being in possession of the truth
on this kind of issue. The first we have just equipped her with:
we stipulated that p is true, and that she is disposed to tell us that
p. Does she have the second? Or rather, since she might have the
second in some other way (perhaps she saw that the cat was on
the mat before Fred told her it was): does she have the second by
virtue of having been told that p by Fred? The answer to that, at
this stage, is clearly that she does not; all we have yet posited is
that what she will tell us is in fact the same as what Fred told her
nothing so far about any connection between the one event and
the other. But before we go further there is one point we ought
to know how to cope with. At least it is true of Mabel that what
she will tell us is the same as what she was told by a person who
was telling her the truth and was very likely to do so. Now that
property is surely one which correlates excellently with her now
telling us the truth; and in that case is it not a perfectly good indi-
cator-property, and one which she possesses?
In a sense it is, but the sense is completely insubstantial. From
the point of view of someone who may now use Mabel as an inform-
ant, let us say ourselves, it has no content. For since we do not
think that Mabel's opinion depends on Fred's, we are in no position
to judge that it is the same as Fred's unless we know independently
what Fred's opinion was. And since we shall in any case have to
believe that whatever Fred held on the subject whether p is very
likely to have been true, it is then really Fred whom we shall be
136 S E C T . XV

using as our informant on this question and not Mabel at all. Earlier,
in Section III, we considered the case of someone who holds a true
belief that p but is useless as an informant because we can only
tell that his belief is true if we already know that/) ourselves. What
we have here is the same thing, but just as it were one link further
down the chain of communication. What is needed is something
or other about Mabel which enables us to judge what it was that
Fred told her from what she tells us, something which we can detect
in her without first knowing what Fred said; that is to say: at the
very least some kind of reaction to Fred's assertion. Perhaps we
have here the beginnings of the intuition that if Fred's knowledge
is to be transmitted to Mabel, so that she too now knows that p,
some kind of 'uptake' on her part is required.
What kind? We have implied that it will have to be at least enough
for her to be disposed, as a result of being told that p by Fred,
to tell us that p rather than that not-p, or nothing at all. Shouldn't
we consider the freakish case in which we know that Mabel systema-
tically distorts this kind of information, and so can gather from
what she says that/) is true although she actually tells us something
else? But we have considered this sort of thing before, and we should
make the same response now that we made in Section V. Such exam-
ples may at best succeed in showing that a certain condition is not
strictly necessary, a point which, though worth noticing, shouldn't
affect our conception of the prototypical case. But does this one
even do that? If we have to work our way back to the proposition
that p from Mabel's assertion that q it is at best doubtful whether
she really is functioning as an informant as opposed to a source
of information, and hence whether there is any real informant in
this story apart from Fred. We can therefore confirm the implication:
Mabel must react to Fred by acquiring the disposition to tell others
what he told her.
To get a little nearer completeness, we may also spend a moment
on the case in which the disposition acquired is simply to repeat
Fred's words on request, maybe uncomprehendmgly, as one might
parrot a message in an unknown tongue (a reaction even more mini-
mal than what I just now called the absolute minimum). Here any
hesitation one may have felt over the previous case is surely absent:
the only informant in the piece is Fred, and Mabel is no more func-
tioning as an informant than is a page of a book, or a strip of magnetic
tape on which Fred's speech is recorded. She must, it seems, be
SECT. XV 137

disposed genuinely to tell us whetherp, not just to mimic the sounds

which Fred made in response to that question.
Must she also believe what he told her, and so offer it to us as
a belief of her own? That seems to have been Michael Welbourne's
intuition about the conditions for the transmission of knowledge:
'I told him who she was but he didn't believe me' is a correct and sensible
report of the frustration of the speaker's act of commonmg knowledge by
the hearer's refusal to accept it. l
Again, we find ourselves somewhere we have been before. Whether
a good informant must believe what he tells the inquirer, like the
question whether a knower must believe what he knows, is some-
thing that needs to be judged in the light of Colin Radford's examples
of the French Canadian and the hesitant librarian. 2 We find our-
selves in the penumbra of the conceptual practice: virtually all cases
in which Mabel tells us that p will be cases in which she believes
that/? and would not tell us unless she did believe it. In our attempt
to understand the practice we have to give that fact all the weight
it deserveswithout ignoring the word 'virtually'. We can say, if
we want to, that Mabel must believe what Fred told her, so long
as that 'must' heralds not a logically necessary condition but a
description of the prototypical caseotherwise we invite an
expedition into the penumbra, where there is no reason to expect
that anything decisive can be made to happen.
So, with that cautionary note loudly sounded, we can allow our-
selves to say that Mabel must believe what Fred told her, because
Fred told her. Is that enough for her to be a good informant on
the matter in hand? To put it another way: is that much 'uptake'
sufficient for the property of being a good informant to be transmit-
ted? It would notso at any rate intuits Welbourne, and my intui-
tions concurbe sufficient for the transmission of knowledge. For
suppose she mistakenly took Fred to be attempting a double bluff,
that is to say, trying to get her to believe the opposite of what
he said by trying to get her to believe that he was lying, when
he knew perfectly well that he was telling the truth. Then she would
end up believing what he told her, and believing it because that
was what he told her (for had he told her that not-p, then not-^
is what she would have believed)but would she know that />?

1 2
M. Welbourne, p. 93. Section V, above, and C. Radford.
138 S E C T . XV

Many will surely feel that she wouldn't. Mabel is suffering from
the notorious false lemma, for if she were brought to realise that
Fred isn't double bluffing, what would she then believe? Would
she believe that not-p, because she would then think that he was
merely (single) bluffing? Or that/?, because she would then think
that he wasn't bluffing at all? Or maybe nothing, because she then
wouldn't know what to think he was up to? So far as our story
goes, it could be any of these. Lucky old Mabel then, to have got
the answer right after all.
Consider now the parallel question as to whether Mabel is a good
informant. We, who want a good informant and are hoping that
Mabel will do the job, believe that she is telling us what Fred told
her; further that Fred was potentially a good informant, and if sincere
in what he said actually was one. Probably at this stage -we rate
her chances of being right very high, but then we hear that she
thinks that Fred was double-bluffing, and that in fact he wasn't.
How are we now to assess the likelihood that what she tells us
is the truth? We have to assess the relative likelihoods of two possibi-
lities: that he was only bluffingin which case she will be wrong
and that he wasn't bluffing at allin which case she will be right.
And all we have to go on is that something or other made her think
he was double bluffing. We are all at sea.
Suppose, however, that we are in a position to make some such
estimate. Suppose we become virtually certain, and rightly, that
Fred was being perfectly sincere. Now -we know, since we can work
out the effect of her view that he was double-bluffing, that Mabel
is just as likely to be right as Fred was, which means very likely
indeed. But now we are in the position of knowing much more
about the situation than Mabel does, and it is very doubtful whether
we are using her as an informant as to whether p. What we are
doing, rather, is using her as an informant (or maybe as a source
of information) to find out what it was that Fred told her. On that
quite different question she is a good informantthough she didn't
get the title by inheritance from Fred, but by using her own eyes
and ears. And once we have that piece of information, we can do
the rest, reliably. She could not; from what we know about her,
her route is altogether too hazardous.
That, then, is why it is not enough for Mabel to believe what
Fred tells her, not even if she believes it because it is what he told
her. What we want, it now seems, is that Fred should have been
S E C T . XV 139

sincere, and that Mabel should believe him to have been sincere
and believe what he told her (i.e. thatp) for that reason. This appears
to be what Welbourne calls 'believing Fred', and holds to be necess-
ary for the transmission of knowledge. And as far as Mabel's contri-
bution is concerned, he takes it to be sufficient: 'All that is required
of a listener who understands a knowledgeable teller if the knowledge
is to be successfully transmitted to him is that he believe the teller'.3
Necessary it may be, but I doubt whether it is sufficient (even sup-
posing that Fred satisfies all required conditions on the teller), for
we can repeat the recipe used at the previous stage and insert another
'deviant' link: let Mabel's (true) belief that Fred was being sincere
(and was 'speaking from knowledge' etc.) be reached by some weird
route involving a whole lot of false beliefs on her part; so structure
the story, if you will, that only an amazing series of coincidences
prevented her from discovering their falsity and thereupon taking
Fred to be lying, or falling into complete confusion as to whether
he was lying or not. Then whether the exchange leaves her knowing
what Fred knows becomes as doubtful as ever.
Perhaps, however, it isn't Welbourne's intention to define 'believ-
ing the speaker'; perhaps he is offering it as an undefined intuitive
notion. But if that is to succeed, then on my diagnosis it would
have to be just an unrecognised attempt to exclude all deviance from
the route by which Mabel arrives at her belief, to ensure by covert
stipulation that she is not prey to any false lemma. It would be
preferable, because ultimately more illuminating, not to accept it
as primitive but rather to ask what was involved. Then we would
come to see what work it was really doing.
M. Welbourne, p. 5.

We have concentrated on the form 'knows whether p', where p

is some proposition. But it hardly needs to be said that there are
other locutions using the verb 'know'; particularly common are
'knows how to A', where 'A' stands for some verb or verb-phrase,
such as 'swim' or 'ride a bicycle', and that in which 'know' takes
a direct object, 'knows X'. Very frequently 'X' here designates a
person, though by no means always: there are people who know
the law, and those who know their Mingnot to mention those
who know London, Japanese, and their onions.
It would be nice if our account of 'know' in the prepositional
context could be coaxed into throwing some light on these locutions
as well. Otherwise we shall be left with an apparent ambiguity,
a term whose semantics have (at least) two unrelated explanations.
Obviously there are many such terms, but it would be uncomfortable
to have to number 'know' amongst them. Ambiguity, in the sense
in which homonyms are ambiguous, doesn't feel to be the right
explanation here. There is, of course, one contrary indication for
the locutions 'know that (or whether)/;' and 'know X': very many
languages translate 'know' differently in the two casesas regards
Europe, at least, English is very much in the minority. 1 This is
just what one normally finds in the case of homonyms: they tend
to be language-specific, and their ambiguity not reproducible in
translation. But a further look at some of the linguistic facts nullifies
any such suggestion. For instance: the German Erkenntnis, clearly
related to the verb kennen (which is of course the form taking the
direct object), is quite happily used to include 'know thatp' cases;
and Erkenntmstheorie (the theory of knowledge) is very far from
restricting itself to the direct object version of the phenomenon.
Similarly, Polish has wiedziec (which takes a that-clause) and znac
(direct object) and derives its expression for the theory of knowledge
in general from the latter: teona poznania. Hungarian has tudni
Perhaps a minority of one, so far as I have yet been able to discover.

(hogy) = know (that) and ismerni + direct object. Again, the word
for the theory of knowledge derives from the second: ismeretelmelet.
Nor is any restriction to the direct object form implied in the French
equivalent theorie de la connaissance, although connaitre is used
with the direct object construction. The theory of banking, which
is understood to some degree by both economists and aircraft
designers, is quite a different businesstwo quite different busi-
nesses, in fact.
It should be added that when we go outside Europe the outlook
begins to change: English no longer looks like the odd man out.
In the handful of African and Asian languages on which I could
find a good informant, all without exception turned out; to share
the English pattern: they commonly translate 'know (that)' and
'know (somebody)' by the same word. In view of this there is no
question of shelving the issue, saying that it is altogether too paro-
chial a linguistic fact to expect our style of pragmatic synthesis to
account for it, rather as if one had asked State of Nature Theory
in political philosophy to explain some idiosyncracy found in only
half a dozen of the world's constitutions. On the contrary, it is
something very widespread. The obvious exceptions ('obvious', that
is, when judged from the western edge of Europe), as I have just
said, are only dubiously such; to use them as a smoke-screen would
be to duck a pressing issue.
There are, however, a couple of methodological points on which
we ought to dwell a little, since they will be important for the subse-
quent section as well as this one. My type of explanation is best
suited to explaining why a certain concept or linguistic practice
exists, why languages need expressions to fit certain pragmatically
created slots, so to speak. Now if we regularly find the same word
filling two slots it is fair to conclude that the slots must be related,
and to try to specify the relationship. But we must be clear: there
is no commitment to the view that, if two slots are related, one
will find the same expression in each. Be the difference only gramma-
tical (here a 'that' clause, there a direct object for instance), no practi-
cal explication, however impressive an example of its kind, can yield
the result that all languages must use the same word in both places.
If native speakers can make the distinction, in any way at all, then
they could use two words to mark it. It may be said that if the
roles are conceptually very similar the verbal distinction will in due
course evolve away. Perhaps, but this has the unsteady feel of a

piece of armchair linguistics. That logically anomalous features of

natural languages don't persist sounds like a guess, and a philoso-
pher's guess to boot, and on top of that a philosopher who has
forgotten how often other philosophers have used alleged logical
anomalies in grammar as an explanation of conceptual illusion. The
fact that 'know that' and 'know someone' is an instance of a wide-
spread linguistic pattern obliges us to look for a connection; but
finding one wouldn't commit us to the view that the pattern must
be universally observed, and it shouldn't surprise or distress us to
find exceptions. We shall find the same with 'know how to'.
A second point, at which I have already hinted, is this: it is positi-
vely undesirable that our method be used to explain linguistic pheno-
mena specific to a few individual languages. It will help here to
remind ourselves of our starting-point and motivation. The idea
was that since 'know' expresses a concept which, give or take a
few niceties of usage, is found in all languages, we accordingly wanted
to explain how its central core arises out of very general conditions
of life such as must affect any language-using community, perhaps
any community at all. There is no such motivation for trying to
capture in our explanation semantic nuances or facts of grammar
found in only a few individual languages. On the contrary, there
is a strong motive for not doing so: if we picked up anything specific
to one language, that would imply that this language had somehow
remained more faithful to the origins of the concept of knowledge
than had others. The implication would smack of chauvinism (the
language in question is almost certain to be one of the few well
known to the investigator); and besides that it would raise the ques-
tion why this language had stuck so closely to those origins when
others had not. And since some of those others would be languages
obviously related to the favoured language, and some might even
be amongst its indisputable ancestors, the question would very likely
be totally intractable. On all counts, therefore, better to steer away.
Nothing of this sort seems to apply, however, to the construction
'knows + direct object', which is clearly extremely widespreadso
back to business. At first glance it might look as if the obligation
to exhibit some kind of connection between knowing Fred and
knowing that it is raining was the simplest of things to discharge.
We are frequently in the position of wanting information, not on
some particular point (whether p), but on some particular subject-
matter. We need an informant who knows a lot about it, whatever

it iswithout, perhaps, wanting to specify at this stage exactly what

questions we need him to answer. So a way of talking arises suitable
to express this need: we are looking for someone who 'knows X',
where X is the subject-matter in question. Someone who 'knows
the law', then, is simply someone who can give us the right answer
to a lot of legal questions; and someone who knows Fred is simply
someone who can tell us a lot of truths about Fred. Easy!
We must not underestimate the force of that way of looking at
the direct-object construction in general; it can actually take us quite
a long way towards understanding its relationship to the preposi-
tional use. But there are certain particular cases that resist, and one
of the most important is precisely that in which the direct object
of 'know' designates a person. I might, calling on resources of musi-
cological scholarship (real and imaginary) come to know more about
Schubert than anyone else, maybe even more than he knew himself.
But however much I knew I still wouldn't know Schubertfor us
nowadays that takes time-travel, not scholarship. In looking for
a good informant about Schubert I am not (fortunately) looking
for someone who knew Schubert; and if I did perchance come across
a Viennese bicentenarian who knew Schubert he might well not
be as good an informant as some expert who didn't.
So it looks as if knowing Fred cannot immediately and easily
be linked with having information about Fred. It is certainly not
equivalent to knowing a lot about him. The tendency of the discus-
sion might make one equate it rather to standing, or having stood,
in a perceptual relationship to Fred himself, having experienced
Fred's person, if one may so put it. And this would fit very badly
with the original suggestion that the concept of knowledge is unitary
and arises in all its grammatical forms from the needs of an inquirer
looking for a suitable informant.
Nevertheless, on closer examination of its logic, 'knows Fred'
may not be quite as recalcitrant as at first appears. To start with,
let us remind ourselves that the 'knows X' locution does not always
resist the model of the informant, as is shown by bur example of
knowing the law. Rather it is certain values of X,, for instance names
of people or places, that cause the trouble, and this may be taken
to suggest that what we are after here is something connected with
particular types of objectit might be that for particular types of
object, pre-eminently persons, there are particular areas of infor-
mation that we are especially interested in, and that possessing infor-

mation in these areas is hardly ever found except in those who have
frequently been sensorily acquainted with them. That would bridge
the gap between 'knowing Fred' and having information (of this
special category) about him, and it would bridge it in a way which
allows us to assimilate another fact of usage: that knowing Fred
comes in degrees. For we may know Fred very well, quite well,
or hardly at all, so to account for the locution we need some quantity
of which someone can have a lot, or a bit, or very little, and if
knowing Fred does after all have to do with having information
about him, then how much information one has will fit this bill
very nicely. The most that our earlier arguments have shown, then,
is that this range of information is not correctly specified if we just
call it, with complete generality, 'information about Fred'.
Admittedly, there is an alternative way of accounting for these
facts which is less favourable to our project. It says that for certain
values of X 'knowing X' means being from time to time sensorily
related to X, perceiving X, or perhaps more generally 'being in
the company of A". Thus for non-perceptual objects, or more gener-
ally non-locatable objects, such as the law, knowing them is indeed
to be understood in terms of possessing information about them.
But for some objects, especially persons, knowing them is primarily
a matter of being together with them, and related to having infor-
mation about them only because and in so far as that is the normal
outcome of being in their company. And being in someone's com-
pany has the required property of being (roughly) quantifiable: one
may spend a lot, not much, or hardly any time with Fred.
It is gratifying, however, to see that this account fails decisively.
Knowing Fred, whatever it may be, is not something that rises and
falls with the length of time the subject is together with, or experienc-
ing, Fred. I may have spent hours with that fellow-commuter, yet
scarcely know him at all. On the other hand there are people in
whose proximity I haven't spent anything like so long and yet know
(clearly because of the manner of the contact with them) quite well.
The refutation of the 'acquaintance' theory doesn't prove that the
informational approach must be correct, but when one thinks of
the sort of example that refutes the former one cannot but incline
to the view that the latter is a very strong candidate: surely the
key lies in how much, and the kind of, information about Fred
that the subject's contact with him has yielded?
The impression is reinforced when one remembers that lack of

information about someone can certainly tell against a claim to know

them. There is no one question about them failure to answer which
is conclusive, but as evidence of ignorance builds up (I don't know
what he does for a living, whether he has any brothers or sisters,
where he lives, how he likes to spend his spare time, and so on)
my claim to know Fred will be gradually whittled away. And, signifi-
cantly, no amount of evidence of sensory confrontation with him,
the ability for instance to give the most detailed description of what
he looks like or the clothes he usually wears, will reinstate it. 'How
peculiar', it will be said, 'to know that much about Fred's appearance
when he hardly seems to know Fred at all'.
Perhaps our problem should therefore be this: supposing that
the usage 'knows Fred' is at core to be understood on the same
kind of informational model that we have proposed for 'knows that
p', can we explain the feeling that some degree of acquaintance with
or experience of Fred is a necessary condition for it, and that without
it no quantity of information about Fred is sufficient?
The best chances for a solution must surely lie along the following
lines: the nature of life in society makes certain items of information
about people of particular interest to us. We want to know what
Fred may be able to do for us, whether he will be willing to do
it, how we may best approach himjust to mention a few of the
more self-interested of our concerns. Many of these will be the sort
of things that are not apparent to casual, or even careful, sensory
inspection of the person in question. But there often are people
(other than Fred himself) who are good informants on these matters,
and virtually always they are people who have spent quite a lot
of time in Fred's company. It is true that one can imagine a good
informant who has notas in our above example of the world's
greatest expert on Schubertbut it is notable that our example
was imaginary, and one can be sure that if real examples exist
they are very rare indeed. There may be persons who have corres-
ponded with each other long and intimately without ever having
met; no doubt a small number of amateur radio 'hams' are in
this position. And it is significant that if one of them were asked
whether he knew the other or not he might well have difficulty
deciding what to say. At least he would not be likely to give
a quick and firm 'No' just because they had never actually come
face to face.
There is, in other words, information about Fred which, with

very few exceptions, only people who are literally 'acquainted' with
him possess. In seeking a good informant on such topics we therefore
take ourselves to be looking for someone who has spent a good
deal of time withmeaning in much the same place asFred.
Whether that condition is strictly necessary or not is a question
that might be best avoided, but if we must ask it the right answer
is probably 'not'which shouldn't lead us to think that an adequate
account of the concept would not mention it at all. The position
is much like the one we saw earlier, 2 when asking whether it is
a necessary condition of knowing that the knower believe what he
knows, and what we are in danger of getting blocked by is too
much respect for the old analytic format of logical necessity and
sufficiency, with all else consigned to silence.
Apart from terms designating persons, the most common values
of X for which 'knows X' raises the issue of sensory acquaintance
are expressions referring to places'knows London', and the like.
They pose no fresh problems, and in one respect they are less compli-
cated. It is rather easier to state, in general, what sort of information
about London we shall normally be looking for: how to get from
A to B, where certain sorts of facility are to be found. The point
of principle is unaffected, namely that this sort of knowledge is vir-
tually never found in anyone who has not spent a good deal of
time in London. Once we have seen this we need no longer be
too worried about the prospect of establishing close contact between
knowing that p and knowing X, close enough to shed a good deal
of light on the linguistic data.
But still there is cause to go a little further. A small observation
may show that the idea of acquaintance, or perhaps it should be
interaction with, a person, has found its way deeply enough into
the concept of knowledge to have an effect on the outward grammar
of the direct-object form. When we are thinking of knowing things
about Fred, the demise of Fred does not affect the tense in which
we cast the verb: there are people around now who know a lot
about Schubert (died 1828). But were that Viennese bicentenarian
still alive today he would not know Schubert, though it might be
true that he knew him, and quite likely that he would know a lot
about him. The same holds if we substitute the name of a town
for that of a person. True, an example of a town that no longer
Section II, above.

exists may be somewhat harder to light on; but one that has changed
a lot will do just as well.
Our deliberations so far come to this: whilst there is clearly a
heavy informational component to 'knows X', the past-tense pheno-
menon that we have just seen indicates another element, one which-
unlike possession of information about Xceases with the cessation
of X. One thing you can no longer do to X when X has ceased
to be is, obviously enough, perceive it. But there are many other
barriers to perceptual contact. One of the most common is spatial
separation, and we do not change 'I know' to 'I knew' just because
an acquaintance is now far awaywe do that only when we feel
we have 'lost touch'.
As to what this might in essence amount to, I offer a conjecture
which at least has some of the right properties. What 'knows Fred'
suggests (in addition to information) is a capacity which might be
broadly described as the capacity to interact with Fred, and to do
so more smoothly and successfully than is generally the case when
two more or less randomly selected persons come into contact with
each other. (Note, now we are talking of capacities, that the capacity
to recognise Fred isn't of much use here. I don't know Queen Eliza-
beth II, but I would be pretty good at recognising her; probably
some people who do know her would be no better.) Only in the
rarest cases will such a capacity exist in the absence of quite a lot
of information about Fredso the two ideas intertwine.
Compare now two other examples of 'knows X': 'knows Lon-
don', and 'knows German'. The former surely implies an ability
to get about in London, to find places one wants, and so on. It
also implies possession of a good deal of information about London,
but this is not really a separate thing, since an adult human who
can find his way about virtually always knows what he is up to,
and so can present his capacity to himself and others in informational
form, even if only rather crudely. To ask whether capacity or infor-
mation is primary would probably be idle, and in so far as there
are pointers one way or the other they are variable and context-
dependent. In 'knows German' the emphasis is slightly different:
it seems far more closely connected to knowing how to do something
than to knowing whether something is the case, much more like
an ability to act than a capacity to inform, though this may well
be just because people are on the whole less skilled at articulating
their linguistic than their geographical abilities. It is worth noting

that there are languages which commonly use the word for 'can'
rather than 'know' in this connection: 'er kann Deutsch' is thoro-
ughly colloquial for 'he knows German', and Welsh has a very similar
The form 'knows X' can therefore sometimes imply (or maybe
even just straightforwardly affirm) capacities to act in certain ways.
Given that, it cannot be wild to conjecture that this factor may
be in play when X is the name of a person. Persons enter into our
lives in rather a lot of ways, so that we could hardly be expected
to say just what capacities are meant, as one quite easily can where
X names a town, and trivially can where it names a language. But
we already have some idea from remarks I made3 when speaking
of the sort of information which, in our practical dealings with a
person, we might find especially valuable. If we want to enlist Fred's
help a useful person to have on our side will be someone who knows
Fred: he will get a chance to broach the matter, and know how
to go about it, where we wouldn't. If we want to cheer Fred up
(I don't wish to make it sound as if manipulation were all that inter-
ested us), whom would we prefer to send in: a friend or a stranger?
Knowing someone, being acquainted with them and familiar to
them, oils the wheels and has a million uses.
The line pursued in the earlier paragraphs of this section related
'knows Fred' to 'knows that p'. This one relates it to 'know how
to'. We therefore need a view of the relationship between this loc-
ution, which at first sight appears to ascribe a capacity, and the
evidently informational 'knows that p'. Otherwise this attempt to
relate 'knows (direct object)' to both might be thought to split it
incomprehensibly in two. Besides, quite apart from knowing Fred,
enough languages have both 'knows that' and 'knows how to' for
this to be a question in its own right. This section has contained
hints, but now they must be made more explicit.
Above, p. 145.

On the face of it, 'knows how to' is synonymously replaceable by

'can': to say that Fred knows how to swim differs only verbally
from saying that Fred can swim. (Some readers will intuitively feel
that 'synonymous' is too strong, that the difference is a little more
than verbal; they are right, but to avoid speaking of too many things
at once I must ask them to waituntil p. 157.) And 'Fred can swim'
tells us something about what Fred can do, not about his capacity
as an informant. What then of our hypothesis?
We can of course ask someone how to do X, as well as whether
p, and getting the right answer to the former type of question is
often every bit as important to us as the latter. This might appear
to show that the inquirer and his informant have just as good a
place when it comes to the 'how to' construction as they did when
we were thinking about 'knows whether' and 'knows that'. Earlier,
whilst discussing knowing whether/), we took as an example a person
wanting to know the way; and I trust that this was not felt to be
inappropriate or strained, in spite of the fact that such an inquirer
would usually put his question in the form 'Can you tell me how
to get to the Town Hall?' Agreed, he will often receive an answer
in the imperative mood: go on down here to the traffic lights etc.
But an answer in the indicative (it's down there on the right) is
just as normal, and the state he attains, if successful, can perfectly
well be described as having information, knowing that the Town
Hall lies in such and such a direction, or 'knowing where the Town
Hall is'; and he could equally well have asked his question in those
words ('Could you tell me where the Town Hall is?')we often
do. This isn't surprising. There are many things which, under any-
thing like normal circumstances, we can do as soon as we have
a certain piece of informationnothing but ignorance is in the way.
Conversely, there are very many abilities which, except in freakish
circumstances, people do not have unless they also have conscious
access to a certain item of information. So for practical purposes
150 SECT. X V I I

the two locutions can then be treated as equivalents.

This is encouraging, but it shouldn't be taken to dispose of the
problem. When we wonder whether that lady knows how to get
to the Town Hall, then, it is true, we are nearly always wondering
whether she will be able to give us information that will enable
us to get there. But when we agonise over whether a child 'knows
how to get home' we are worried about whether it can find its
way home, not whether it can efficiently direct us there; clearly
it might be able to do the former unfailingly whilst being hopelessly
bad at the latter. What we are apparently concerned with here is
a use of 'knows how to' which is at least nearly synonymous with
'can', and this seems prima facie to have no connection with com-
petence as an informant.
As with the direct object construction, a simple claim of ambiguity
is far too implausible an escape route. Besides, it would leave us
wondering why so many other expressions, none of them with any
feel of ambiguity, consort equally freely with both 'that/?' and 'how
to A': 'learn', 'teach', 'forget', and 'remember'. Furthermore, when
the same ambiguity is found in a number of languages, especially
if they are not all related, pressure builds up against the view that
it is an ambiguity at all. So the claim would give a hostage to linguistic
fact; and (to anticipate) linguistic fact is against it.
What responses are open to us? One might be to protest that
we should not be expected to explain all features of usage; presum-
ably many factors are at work shaping linguistic history. Now if
we are thinking of features of usage specific to one or a few languages
we should uphold the protest, as was argued in the preceding section.
But if whatever is in question is found in almost every language
the position is different. Failure to illuminate one such feature
doesn't, strictly speaking, invalidate a putative explanation of
another. But it would be feeble to hide behind that formal point,
when the near universality of both features cries out for the same
kind of treatment. And since these features concern the usage of
the same word, there are obvious theoretical advantages in a unitary
explanation which makes it understandable that the word should
be the same in each construction. So this protest would really be
just an admission of partial failure, a dive for cover; we should forget
There seem to be three strategies which may offer a satisfactory
account of 'knows how to'. The trouble arose because that expression

appears to have a sense in which it means that the subject can perform
a certain kind of action. One way round the obstacle would be
to disperse it by arguing that this capacity sense, although found,
does not have the empirical status to be a problem: whilst it occurs
in some languages, it is not so widely distributed as to be a proper
subject of the present investigation. A second would be to argue
that, first intuitions notwithstanding, the apparent capacity sense
is really nothing of the kind, but informational after all. The third
is to accept both existence and status of the capacity sense and then,
by exhibiting a natural connection between the ideas of information
and agency, explain how knowing thatp and being able to A come
to attract the same word.
How widespread is 'knows how to'? That is the crucial question
for Strategy One. We are to focus on the obvious way, or ways,
of rendering the English 'knows whether p' in a given language
'obvious ways', to indicate that we are not to be sidetracked into
subtle distinctions of nuance, references to the ways the words used
in the respective language feature in specialised contexts, and so
on. (We are not interested, for example, in whether the expression
that corresponds to 'knows' in the Arabic for 'He knows whether
it rained yesterday' is ever used in what is sometimes called the
'biblical' sense.) Having thus located the analogue(s) of 'know' in
'know whether', we consider the resources standardly available to
a speaker of the given language for describing someone who, as
we would say, can swim, and we look to see whether they (all
or some of them) include the 'knows' component from the first
group of expressions (or something closely related to it). If at least
one of them does, then the given language has the capacity sense
of 'know'.
Notice that the exact grammatical structure, knows how to +
infinitive, is not our real topic. What we are really talking about
are capacity-attributions which use 'know' or its obvious correspon-
dents in other languages. It is that, and not any grammatical isomor-
phism, that we are to concentrate on.'

The nearest German equivalent, for instance, is 'Er weift wie man schwimmt'.
But what this suggests most strongly is that the subject can tell us the theory of
swimming, not that he can swim. Finnish (I am told) also has a construction verbally
close to the English 'knows how to'; but again the import is theoretical. I understand
that Hungarian, on the other hand, uses the word for 'know' with the infinitive
as its standard colloquial way of saying 'can'.
152 SECT. X V I I

So, does the capacity sense of 'know' have status or not? Here
we need the empirical facts. We won't of course be able to get all
the empirical facts; we shall have to be content with (or exhausted
by) the facts about a few particularly prestigious (i.e. well-known)
dead languages and a representative selection of living ones. Living
languages are pretty numerousa few thousand, on any sensible
criterion for distinguishing them from each otherso we must hope
to be representative without having to run our test on more than
a small fraction of them. Our best bet will be to rely on the classifica-
tions of languages agreed among comparative linguists, and hope
to find the resources (which is to say an obliging, educated, expert
speaker) for testing a few members of each of the mam groups.
These tests, so far as I have been able to carry them out, support
the contention that the capacity-ascribing sense, if we can rely on
our intuitive impressions to detect it, is at the least very widespread.
Since the honest policy, and the only one that can lead to a genuinely
robust theory, is to confront directly anything recalcitrant which
looks as if it probably holds, it seems that we should accept that
there is a prima facie capacity-related component of the concept,
and try to explain the fact.
If anyone is determined not to accept it, they will be able to
create a little room for manoeuvre by playing on a methodological
difficulty in the application of this test. I have already hinted at
it: we cannot work solely with formal features of grammar, but
must rely to a considerable extent on the semantic intuitions of native
speakers. There may be languages2 which have expressions that
are close verbal parallels to the English 'know how to' without its
being clear that they have the same meaning. Assume (for the
moment) that the English 'knows how to' is a synonym of 'can';
it might be that in some languages there is a verbal equivalent of
'knows how to' which is semantically informational in content, that
is to say: it is applied only to subjects who can tell us the way
to perform whatever action is in question, or (a very different thing),
the theory of its performance. In that case the mere existence of
the verbal equivalent would not qualify that language for inclusion.
Worse, there may be problems in determining the facts here; it may
not be clear what sense, capacity or informational, a given expression
has. We can experience the sort of difficulties that arise by consider-

And so there are: n. 1 above.

ing the English expression itself, and we can kill two birds with
one stone by doing it in the context of the second of the three possible
The second strategy is to deny that 'knows how to' has a capacity-
sense after all. First appearances notwithstanding, we might argue,
it indicates the possession of information about the performance
of the action specified, perhaps even (we shall soon see that this
is a little more) the capacity to give information about it. There
are two broad types of position that such an argument might aim
at. One would be the view that behind any exercise of ability lurks
the deployment of information: if you can swim, that is always
because you know that such and such is the way to swim, and
can apply that knowledge. When, attributing to you the ability to
swim, we say that you 'know how to swim', the word 'know' is
being used with reference to the essential informational background,
the 'knowing that'; and the change from 'that' to 'how' marks the
fact that we are dealing with information of the type which character-
istically enables a certain kind of action. This position is not far
from the doctrine which Gilbert Ryle3 called, and attacked as,
'intellectualism'we shall shortly look to see whether his attack
leaves any of it standing that might be of interest to us.
That short statement is enough to expose the difficulty I referred
to in the previous paragraph. We wanted to be able to decide whether
or not a given language uses its literal equivalent of 'knows how
to' in a capacity-attributing sense. But, supposing for the moment
the truth of the 'intellectualist' doctrine just outlined, how would
we determine whether the word 'know' really referred to the state
of information, or whether it was primarily a way of attributing
the ability? If a capacity is always based on information, how shall
we decide which of the two the expression 'knows how to' really
describes? It looks as if we may have to accept a commitment to
extensive quantities of semantic theory, something which the project
was designed precisely to avoid.
The proper way to deal with this difficulty, I am convinced, is
simply to bypass the argument and confront the least favourable
alternative head on. Where two descriptions of the situation are
possible, and a decision between them, if not arbitrary, would have
to be grounded deep in the theory of meaning, we should treat

See G. Ryle, ch. II, esp. pp. 30-2.

it either under both, or under the one likely to cause our hypothesis
the most trouble. Thus, going back to our first strategy, we can
see that it will cause the more difficulty the more languages are
deemed to operate a capacity sense of their word for 'know'; so
our principle recommends that we decide in favour of the capacity
sense wherever there are not clear reasons to the contrary. In other
words, we should acknowledge our obligations and commit our-
selves to the third strategy. Otherwise we invite the charge of either
ignoring the linguistic facts or trying to bury them under a pile
of controversial semantic theory.
In any case, the train of thought we were considering is set in
motion by an 'intellectualist' account of knowing how, and in view
of that it may well be asked whether it is worth considering at all.
Hasn't intellectualism been refuted? Ryle, one must admit, gave
strong reason for thinking that, when taken in strict generality as
applying to every capacity, it must be false: it leads to infinite regress.
I accept the argument and its conclusiona little casually perhaps,
but that will do no harm in this context since it doesn't settle the
present issue. For what Ryle showed was that some capacities must
just be things which we can simply do when occasion arises. So
intellectualism (i) can be true of at most nearly all our capacities,
and (ii) cannot be supported by any general argument to the effect
that capacity entails information. Goodbut suppose that it were
true of nearly all our capacities, and that that belief rested on common
observation rather than any grasp of logical connection. That would
be compatible with all that the argument from regress proves, but
it would still be quite enough to explain how a concept which origi-
nated in the need for good informants should grow to encompass
abilities as well.
But is it true, even in this weakened form? More exactly, does
being able to tell us how to A go with being able to A, in our
experience, in a sufficient proportion of cases for there to be any
plausibility in the claim that the concept of knowledge spreads from
the former to the latter precisely because of such concomitance?
That must be very doubtful. Remember that when we speak of infor-
mation we are speaking of being a good informant, not of information
in the sense in which cognitive psychology might use the word.
It is clear that the child who can unfailingly find the way home
must in that sense have a great deal of stored information, and the
same holds of anyone who can swim, ride a bicycle, or do anything
SECT. X V I I 155

else; but just as clearly, that is no guarantee of being able to produce

it in a form that an inquirer can make use of, nor of oneself having
access to the information in the form of conscious belief, nor even
of having any practicable way of gaining such access. Having infor-
mation, in this sense, has nothing to do with being a good informant,
even potentially.
We must not, therefore, by conflating these senses of 'infor-
mation', allow ourselves to exaggerate the degree to which capacity
as agent goes along with capacity as informant. We should also
beware of another factor which might lead us to exaggerate it. Any
reasonably articulate person, if he can do something or other, will
in many cases (not all) be able to tell us something of how it is
done. But it may be that he does that by observing himself doing
it, and then describing what he observes; in such cases the connection
between having the ability and being a (more or less) good informant
about how to do it is a wholly external one. If he saw someone
else doing it he might then be just as good at telling us how it
was done, in certain cases he might even be better. Now this type
of case, though it does produce a certain degree of concomitance
between having an ability and having information, we really ought
to discountit does little to explain why there should be a linguistic
connection between being able to say how something is done, and
being able to do it oneself. It is on a par with the fact that someone
who has a dog is likely to be able to tell us at least a little about
dogs; that hasn't generated a linguistic link between owning an X
and knowing about Xsnor would anyone expect it to have done
The hope of the second strategy was to find grounds for denying
that 'knowing how' really has a capacity-ascribing sense. Our admit-
tedly inconclusive discussion has by its very inconclusiveness
pointed us towards the third strategy: that of accepting such a sense,
and then arguing that knowing how to A lies, in so many cases,
so close to being a good informant about .A-ing as to make it perfectly
comprehensible that a concept which has its seat in the idea of infor-
mation should include the possession of capacities to act as well.
But in pursuing Strategy Three we shall find a little hint of Two
pushing its way in. 'Know how to' is indeed closely related to 'can',
but not so closely as to justify the assertion of synonymy; and prag-
matic reasons can be found to back up the common intuition men-
tioned at the beginning of this section, that 'knows how to' is at
156 SECT. X V I I

any rate not a pure ascription of capacity.

An underlying connection which can account for the overlap in
usage is, in fact, not so hard to find. We may start with the obvious
point that human beings need both true beliefs and capacities to
act, since every action calls for both. The inquirer seeks a true belief
on the question whether p; the apprentice, as we may call him,
seeks the capacity to do A. His purposes may be furthered either
by someone who tells him, or by someone who shows him, how
to do A. So we may consider three cases:
(a) The inquirer, who wants someone to tell him whether^)
(b) The apprentice, who wants either (i) someone to tell him how
to do A, or (ii) someone to show him how to do A.
What we want, as apprentices, is to be able to do A ourselves.
That being so, someone who can show us how to do it will be
just as valuable as someone who tells us, so long as that has the
same effect: that we can then do it. And of course people who can
show us how to do it are invariably people who, at least to some
degree, can do it themselves. We want to be able to do A, and
in many cases we will be equally well served by an informant, who
tells us that such-and-such is the way to do it, and an instructor,
who shows us how to do it by doing it himself in front of us. And
if we are equally well served by either, that will be a factor tending
to encourage the use of the same term in both cases. When nothing
much turns on the distinction there can be no surprise if we employ
the same vocabulary; that is just what is suggested by the idea of
a general term.
So the three cases begin to huddle together. There are obvious
affinities between (a) and ()(i), both of which involve the linguistic
transfer of prepositional information. And we have already observed
that there are affinities between (b)(i) and (b)(ii): for a start, they
are both widely used and effective means to the same end. On top
of that, they are not always even distinguishable, for showing and
telling may merge, and often do: 'Hold the two needles together,
do you see, then take the wool, here, and pass it round one needle,
so, and then take it between the needles like this . . . "
Other factors help to bind ()(i) and ()(ii). There is a class of
cases in which telling and showing, knowing whether and knowing
how, informing and demonstrating, merge; this time not as a con-
tingent matter of normal practice, as in the knitting example, but
SECT. X V I I 157

as a matter of principle. These occur when doing the relevant A

involves giving some verbal or at least symbolic performance, when
to do A is to articulate some propositional information. Knowing
how to prove Pythagoras' theorem is hardly distinguishable from
knowing that it is proved like this: and there follows a sequence
of propositional statements; telling and showing the pupil how to
prove the theorem scarcely differ. It may be thought that such cases
are rather too specialised to make a significant contribution to the
shaping of usage. But that is very doubtful, since from one perspec-
tive a huge class of cases turns out to be of this kind. Suppose we
think of the informant as doing something, namely answering the
question 'whether />?', and doing it correctly. There isn't anything
contorted about this: answering a question is really a fairly central
case of an action, and perfectly naturally seen as such. There is
an intended effect, a choice of means, the production of the effect
by controlled bodily movements. One may choose the wrong end
affirm p when one should have denied itand the wrong means
mispronounce the words, phrase one's answer clumsily or incompre-
hensibly. In the same way one can knock in a nail when one should
have used a screw, knock the nail in badlyon a slant, or only
half wayor miss it altogether. No-one need apologise for thinking
of the informant as performing an action; on the contrary, the onus
rests with anyone who denies it. So if Fred knows that p, if he
is in possession of the information, he knows how to perform a
certain actionhow to answer the question whether p. Thus the
link between possession of information and possession of a capacity
is further strengthened.
We have seen4 that there are languages in which the nearest
verbal equivalent of the 'know how to A' construction carries a
heavy suggestion that what is in question is theoretical, articulable,
knowledge of 'how ^4-ing is done', which need not be accompanied
by any ability actually to do it. And I think it is fair to say that
there is a little-in some cases a very littleof this feeling about
the English expression. We need to distinguish two points:
(1) That 'S knows how to do A' can be true when 'S can do
A' would be false, for instance, if 5 has lost the requisite
physical powers.
(2) That 'S knows how to do A' differs from 'S can do A' in
Seep. 151, n. 1.
158 SECT. X V I I

an implication that S has or can call up some degree of reflective

grasp of the way to do A.

Both of these features are easily understood if we think of them

in the light of the apprentice's situation. He wants to be able to
do A, and is indifferent between (b)(i) and (b)(\i) as ways of bringing
this about, or at least of bringing it nearer. Now it is generally
true that someone who can satisfy his demands by the first method,
who can say how to do it in sufficient detail and in such a way
as significantly to help him towards being able to do it, will be
someone who can do it himself. And anyone who can help him
by the second methodthat of demonstrating itwill obviously
be someone who can himself do A. Hence the feeling that the appren-
tice must be looking for someone who can do A, and hence, on
my hypothesis, the feeling that 'S knows how to do A' entails 'S
can do A'. But since there are casesthe most frequent is probably
that of the ageing expert who has lost the necessary physical
powersin which someone who cannot use (b)(n) may still be a
most effective operator of ()(i), the apprentice's turn can be served
by someone who cannot do A. Hence the existence of counter-
examples to the thesis that 'knows how' entails 'can'. And this
explains the existence of the feature (1). The situation is very much
like that encountered when one asks whether'S knows that/?' entails
'S believes that/)'. Believing is indeed central to knowing, but still
one can find plausible counterexamples against the entailment claim.
We saw that our method could explain that fact, and it can also
explain the similar fact about the relationship between 'knows how
to' and 'can'.
We can also explain feature (2). In the case of many activities,
just performing them under the eye of the apprentice will not be
very effective. Far more effective will be a self-conscious, reflective
demonstration in which the teacher draws attention to the essential
features of the performance, perhaps doing some of them more
slowly, or exaggerating them slightly. To do that calls, obviously,
for a little more on the teacher's part than the bare ability to do
A; it calls for a degree of conscious understanding of the process.
And that is just what feature (2) consists in.
It might be thought that this whole idea faces a tricky objection.
Being a good informant, as we have seen, means more than just
being right; in addition to that the good informant must possess
SECT. X V I I 159

some characteristic that makes him recognisable as such and supports

confidence in his information. That was why the concept we con-
structed from the idea of the inquirer's search matched popular ana-
lyses of the concept of knowledge in calling for true belief plus
satisfaction of some further condition. Now we are trying, by com-
paring the situation and needs of the apprentice to those of the
inquirer, to assimilate 'knowing how' to being a good informant.
So a person who knows how to do A ought in the first instance
to be someone who (1) can do A, (2) is prepared to display ^4-ing,
and (3) satisfies some further condition which advertises (to the suffi-
ciently discerning) his fulfilment of (1) and (2); that would give us
the desired parallel with the good informant, who can tell us the
truth about p, is prepared to do so, and has some property which
reliably indicates it. But not so, surely: knowing how to do A is
just concerned with (1), being able to do A, and not at all with
(2) and (3). The parallel collapses.
The objection is, I believe, confused. As regards condition (2),
we should remember that there was the same problem over knowing
whether p. A good informant should be prepared to assert that p,
or that not-p, if that is his belief; whereas a knower (remember
Luigi) may not be. There we had recourse to the idea of the growing
objectivisation of a concept, and the same device will be available
here; there is no asymmetry at this point. But still there may be
asymmetry at (3). Doesn't the difficulty remain that knowing how
to doyl is concerned with (1), perhaps related to (2), but has nothing
to do with (3)?
It is not straightforwardly true, howeveras the objection
impliedthat knowing how to A has to do with (1) and not (3).
That depends on how (1) is to be understood. There is after all
a minimal conception of 'can', the one enshrined in the old logical
dictum ab esse ad posse valet consequentia, in which I can do A
if I bring it off, even though my doing so was a monstrous fluke
which I am most unlikely ever to repeat. One who says that if
you can do A then you know how to do A is certainly not thinking
of this sense of 'can', but of something much more substantial: that
there is something about you which makes you at least fairly reliable
when it comes to /1-ing. And what: that means, it will at once be
seen, is that if 'can' be taken substantially then the requirement
(3) is already built in; whereas if it be taken minimally we need
the explicit addition of (3) to reach anything recogmsably like know-
160 SECT. X V I I

ing how. The supposed asymmetry disappears.

It would in fact be both surprising and perplexing if there were
any serious failure of parallel between knowing how and knowing
whether. For if answering an inquirer's question is an action, and
as such something that can be done successfully and unsuccessfully,
in the right way and in the wrong way, then every case of knowing
whether just is a certain type of knowing how; that is no less true
just because it is a rather specialised case of it.
Possibly certain asymmetries do remain, though if so they are
not such as to do any damage to our thesis; on the contrary, they
are comfortably explicable in terms of it. Of the many types of
action which one might wish to have done, or see demonstrated,
the vast majority virtually never happen by accident. That means
that a single performance given to order often has, de facto, the
force of a proof that the performer can perform reliably. In contrast,
there are many questions that can be answered correctly by accident,
so that getting the answer right once only rarely establishes much
likelihood that the informant will be right in future on that type
of question. This point affects, however, not the parallel between
the conditions for knowing how and those for knowing that, but
only the ways in which we can establish that they are satisfied in
a given case. A second asymmetry has just the same effect: someone
who cannot himself do A may have no difficulty in recognising just
by observing his performance that someone else is doing it; whereas
someone who does not yet know whether p cannot tell just from
hearing it affirmed that p (or not-p, as the case may be) that he
has been told the truth. This means that whilst in the case of knowing
that a great deal of strain falls on features of the informant which
make it likely that his answer will be true, in the case of knowing
how the corresponding facts about the agent seem of little import-
ance. The cases in which we have to know that somebody is good
at .A-ing in order to know that yl-ing is what he is now doing are
by comparison rather rare. Mostly, therefore, if someone comes
along and does A those who see it will at once, and very reasonably,
take him to know how to do A. That does not mean that nothing
more than his actually doing A is involved; but it may do something
to explain the tendency to think that nothing more is involved,
and to supposenotice that this is in effect the point from which
our objection beganthat 'knowts how' requires nothing more than
SECT. X V I I 161

We should not therefore be worried by these apparent disanalo-

gies. They only concern differences in the epistemology of the two
notions: how we tell whether a subject knows whether/?, and knows
how to do A. Far more impressive is the close similarity of structure
between knowing how and knowing that, something which the dis-
analogies leave untouched. But we should remember that in these
last few pages we have only been countering an objection, not making
a part of our positive case. The fact that the two notions both involve
a success-clause plus a clause indicating that the success is no accident
is encouraging, but it does not of itself go far towards explaining
why the constructions by which they are expressed should use the
same word. For the positive reasons as to why this should happen
in so many languages we look back to the materials assembled in
the earlier paragraphs of this section.

A part of Section XII was directed to material that Unger published

in 1976 in his book Ignorancea Case for Scepticism. Since then
there has appeared his Philosophical Relativity (1984) in which he
expresses views incompatible with those of the earlier work. It would
therefore seem only fair to ask how his new position relates to my
remarks, both critical and positive; besides, what he says has great
intrinsic interest, so that it is not just on grounds of fairness that
it demands our attention.
In his second book Unger introduces an innovatory distinction
between Tnvariantist' and 'Contextualist' semantics.' To be an
invariantist with respect to the semantics of a particular term is to
hold that there is a single standard for its correct use which applies
in all contexts in which it is or may be employed. A contextualist,
on the other hand, holds that the standard for its application changes
with the context of utterance. Since each regards these standards
as determining what the word means, this amounts to saying that
it has one meaning (and one set of truth-conditions) for the invarian-
tist, many (dependent upon context) for the contextualist.
Unger's next move is to argue that there is no way of settling
the question whether the semantics of a given expression are invariant
or context-dependent. The hard data of semantic theory are the
utterances of speakers and the reactions of hearers, and these under-
determine the (supposedly) factual matter at issue between invarian-
tist and contextualist. The difference between the two types of theory
lies in where each locates the complexity which must undoubtedly
characterise the process by which competent speakers pass from
situation to appropriate utterance, and from utterance to appropriate
reaction. The contextualist makes semantic understanding, the grasp
of what has literally been said, highly complex: we have to work
out just what that utterance means in that context. For the invarian-
tist that stage is comparatively easy: we just have to grasp what
See P. Unger, (4), pp. 6 ff.

the utterance means, a matter to which the context of utterance

is irrelevant. But then, for the invariantist, comes the complexity:
we have to be able to judge just how much, and what kind of,
departure from the one invariant semantic standard is acceptable
in the given context. Each theory has to balance the budget: one
buys simplicity here for complexity there, the other pays complexity
here for simplicity there. How much complexity has to be postulated
may well be an empirically determinate issue, but where it falls,
Unger argues, is not. Neither, therefore, is the debate between invar-
iantist and contextualist, since this is the very point on which it
Now Unger's previous argument to scepticism was based on the
lemma that 'know' is an absolute term; and to regard something
as an absolute term is to ascribe to it invariant semantics. (It was
because he then took it that knowing calls for absolute certainty
than which nothing could be more certainand calls for it invar-
iantly, context-independently, every time it is uttered, that Unger
could pass to the radically sceptical conclusion that virtually every
claim to knowledge is, strictly speaking, false.) But if there is also
a contextualist position about the semantics of 'know', and the issue
between contextualist and invariantist is in principle indeterminate
of outcome, then the absolutist theory of 'know' cannot be determm-
ately right, and the argument to scepticism cannot be determinately
successfulwhich I offer as a polite way of saying that it fails.
I have argued, however, that the absolutist reading of 'know'
is mistaken: were it correct, I suggested, no debate about scepticism
of the warmth which is actually found could exist. But my reason
for rejecting the absolutist reading is for the moment immaterial.
The question is rather, whether just by rejecting it I am not in
conflict with Unger, who maintains (does he not?) that it cannot
be definitively rejected: that would be definitively to reject the invar-
iantist view of its semantics. Unger's relativism does in a sense
involve the rejection of absolutism, but not in this way. He does
not so much reject it as its claim to be determinately better than
any of its rivals.
There is, or at any rate should be, no conflict here. Unfortunately
Unger sometimes writes as if an invariantist view had to be absolutist
(which, if it were so, would make the two terms co-extensive), but
his reasons for saying this, if indeed he really wants to say it, are
quite unclear to me. The invariantist says that there is a single,

invariant, context-independent standard for knowledge. That is one

thing; to say that it has to be the toughest standard that is even
in principle possible seems to be quite another. Consider 'X is a
doctor'. An absolutist account of the meaning of that sentence might,
I suppose, be something like this: X is so good at diagnosing and
curing illness that nobody could, even in principle, be any better
at it. I imagine that most of us would want to reject this account
as simply wrong, or at the very least as totally arbitrary and unmoti-
vated by anything in our thought or linguistic practice. But surely
that does not bar us from thinking that 'doctor' needs an invariantist
definition, for instance: a doctor is someone who has passed certain
exams and so acquired certain qualifications. If you have got them
you are a doctor, if you haven't you aren't; context means nothing,
however grateful you might under certain circumstances be for the
attentions of some medically very knowledgeable person who
wasn't, however, a doctor. Invariantism without absolutism cer-
tainly seems to be an option.
Once we realise this we see that the distinction between Invarian-
tism and Contextualism does not bear directly on the truth or falsity
of scepticism. An invariantist doctrine of the semantics of 'know'
might be one which set the invariant standard at some level which
we all reach frequently, so that a high percentage of everyday ascrip-
tions of knowledge turn out to be true and the claims of scepticism
to be based on mere misapprehension. A contextualist, on the other
hand, might be able to argue that certain features of the contexts
of most of our utterances about knowledge set the (variable) stan-
dards at a higher level than we realise, with the result that very
many of these utterances, maybe even whole classes of them, were
actually false; he would award the sceptic a significant victory. Just
how plausible he could make this line appear remains to be seen,
but pending developments it has to be ranked amongst the prima
facie possibilities.
On the second of these points, including the difficulties which
the combined contextuahst-sceptic will face, Unger is quite clear.2
But the first he appears to have missed, for immediately before these
remarks about contextualism he says that 'An invariantist account
of 'know' will directly give the day to scepticism about knowledge,
at least to a fairly extreme form of that view' (original italics). This
Ibid., p. 51 l.iI ff.

implies the equation of an invariantist with an absolutist account,

something which is also implied earlier.3 Now some will read this
earlier passage as more than an implication. This (they will rightly
say) is where Unger introduces the term 'invariantism', so we have
not an implication but a definition: Unger means invariantism to
be the same doctrine as what he in his earlier work called absolutism.
That may be so, though I doubt it. But if it is, I would observe
that there is room for a doctrine which is neither contextualism,
nor that proposed by Dretske which Unger also discusses,4 nor
invariantism as Unger seems to introduce the word, that is, as syn-
onymous with absolutism. This doctrine is what I have up to now
been calling 'invariantism', which seems a good word for it in view
of its content: an invariantist with respect to a certain expression
gives its semantics in terms of a single, context-invariant, standard.
Absolutism is one type of invariantism, but only one type, and
we might therefore declare an absolutist account of some expression
simply false without implying that any invariantist account of it
would be simply false and some contextualist account of it true.
Unger's Semantic Relativism, the idea that for many expressions
there are plausible semantics of both invariantist and contextualist
stamp such that no matter of fact could decide the issue between
them, can still be accepted.
But isn't there still some kind of clash between my procedure
and Semantic Relativism? I have spoken (in Section XI and else-
where) as if the semantics of 'know' were invariant, and the questions
to be answered were simply: what is the invariant standard which
captures them, and why is that the standard we arrive at? Much
use was made, it is true, of the notion of contextually determined
standards, but that was because we started our thought-experiment
from the subjective standpoint of the individual with his own particu-
lar interests and situation in order to see how the process of 'objecti-
visation' leads to a concept whose conditions of application are
intersubjective and far more narrowly circumscribed. Am I not then
found to be maintaining something which can be truemeaning
determinately trueonly if Unger's Semantic Relativism is false?
Again, I think not; the impression of disagreement is superficial.
Readers of Unger's Philosophical Relativity will recall that in the

3 4
Ibid., p. 91.10-12. Ibid., pp. 30-3.

case of any particular utterance there are various things which are
agreed between invariantist and contextualist. They are agreed that
there is a certain thought on which the speaker brings the hearers
to focus; what they disagree on are the roles played, in the process
by which he does this, by the semantics of his words on the one
hand and the pragmatics of the situation on the other. But if they
are agreed on that, there can hardly be any difficulty of principle
for them in agreeing that there may be some expressions which are
characteristically used to fix attention on pretty much the same
thought (here in the sense: the same stringency of standard) in all
contexts of utterance. Whether a particular expression is like this
or not may be contentious; whether 'know' behaves in this way
may be a point of disagreement between myself and Ungerthough
more of that in a moment. But contextualist and invariantist as such
are not committed either to the view that there are, or that there
are not, such expressions.
What can be said is that if some particular expression is agreed
to be of this type, a contextualist account of it will look a somewhat
round about way of reaching the place which an invariantist reaches
in one step. But even if there were simply no hope for the contextua-
list view in such a case there would be no conflict with Semantic
Relativism. For that doctrine, if I understand it correctly, tells us
that where a range of standards is in use the phenomena can be
equally well explained on invariantist and contextualist models. Of
expressions which are not associated with a range but with only
with one, invariant standard, it tells us nothing, leaving us free to
conclude that there the balance tips towards invariantism. But in
any case it is not true that my account of 'know' leaves nothing
that contextualism can usefully do, and that for two reasons.
The first involves an indirect use of the basic contextualist idea.
It is exactly the use that I have already made of it, though not under
that name. What we were considering in Section X (on 'Objectivisa-
tion') was the way in which contexts, constituted by varieties of
need, prospective outcomes, recognitional capacities, levels of ignor-
ance of prevailing conditions, and so on, guide our assessment of
sources of information. And it was a feature of these contexts
principally, the fact that one often does not know what the infor-
mation is needed for, or will later be used for, or what will turn
on its usethat created an obvious role for a concept tied to some
high standard of reliability. So the contextualist has certainly had

his share of the action, even if that action results in a situation for
which, some might think, only the invariantist need apply.
The second point is more straightforward. To say that knowledge
always implies a very high level of reliability is not to say that the
level is always the same. Many everyday claims to knowledge are
allowed to get by although made with thinner support than many
others, which, occurring in different circumstances, are questioned
and even rejected. Looking at the cases in which we are apparently
laxer, the invariantist can say that these knowledge-claims are false
but that we judge according to context what degree of departure
from strict truth is appropriate and acceptable; the contextualist can
say that they may well be true, since context affects the details of
their semantics. On the question which position, if either, is right,
I am happy to follow Unger's powerfully argued recommendation.
My 'practical explication' or 'state of nature' method leads to an
account of the linguistic practice surrounding the word 'know' and
its near relatives; it does not determine how we are to apportion
the underlying mechanics of the practice between invariant semantics
and contextually motivated pragmatics.

Armstrong, D. M., Belief, Truth and Knowledge (Cambridge University

Press, 1973).
Austin, J. L., 'Other Minds', in Philosophical Papers, ed. Urmson and
Warnock (Oxford, Clarendon Press, 1961).
Blackburn, S. W., 'Knowledge, Truth and Reliability', in Proceedings of
the British Academy, lxx(1984), 167-87.
Craig, E. J. (1) 'The Practical Explication of Knowledge', in Proceedings
of the Aristotelian Society, (1986-7), 211-26.
(2) The Mind of God and the Works of Man (Oxford, Clarendon
Press, 1987).
(3) 'Nozick and the Sceptic: the Thumbnail Version', in Analysis,
vol. 49(1989), 161-2.
Descartes, R., Meditations on First Philosophy, in The Philosophical Writings
of Descartes Vol. II trans. Cottingham, Stoothoff and Murdoch (Cam-
bridge University Press, 1984), 3-62.
Dretske, F., 'Conclusive Reasons', m Australasian journal of Philosophy
xlix(1971), 1-22.
Edwards, P., 'Bertrand Russell's Doubts about Induction', in Logic and
Language (First Series), ed. Flew (Oxford, Basil Blackwell, 1960).
Forbes, G., 'Nozick and the Sceptic', in Philosophical Quarterly, xxxiv
(1984), 43-52.
Gettier, E., 'Is Justified True Belief Knowledge?' in Analysis, xxiii (1963),
Goldman, A. (1) 'A Causal Theory of Knowing', in Journal of Philosophy,
Ixiv (1967), 357-72.
(2) 'Discrimination and Perceptual Knowledge', in ibid., Ixxiii (1976),
Grice, H. P. 'Logic and Conversation', appended to Understanding Argu-
ments by R. J. Fogelin (Harcourt Brace Jovanovich, 1978), 329-42.
Harman, G., Thought (Princeton University Press, 1973).
Hintikka, J., Knowledge and Belief (Cornell University Press, 1962).
Hume, D., An Enquiry Concerning Human Understanding, 3rd edn., ed.
Selby-Bigge, rev. Nidditch (Oxford, Clarendon Press, 1975).
Lehrer, K., 'Knowledge, Truth and Evidence', in Analysis, xxv (1964-5),
R E F E R E N C E Ss 169

Lewis, D. K., Counterfactuals (Oxford, Basil Blackwell, 1973).

McGinn, C., 'The Concept of Knowledge', in Midwest Studies in Philoso-
phy, ix(1984), 529-54.
Moore, G. E. 'A Defence of Common Sense', in Philosophical Papers
(London, George Allen and Unwin, 1959), 32-59.
Nozick, R. (I) Anarchy, State and Utopia (Oxford, Basil Blackwell, 1974).
(2) Philosophical Explanations (Oxford, Clarendon Press, 1981).
Pascal, B., Pensees, trans. Krailsheimer (Penguin, 1966).
Pears, D. F., What is Knowledge? (London, George Allen and Unwin,
Plato, Meno, trans. Guthrie (Penguin, 1956).
Radford, C., 'Knowledgeby examples', in Analysis, xxvii (1966-7), 1-11.
Ramsey, F. P., 'Knowledge', in Foundations: Essays in Philosophy, Logic,
Mathematics and Economics, ed. Mellor (London, Routledge and Kegan
Paul, 1978), 126-7.
Ross, A., 'Why do we believe what we are told?' in Ratio, xxviii (1986),
Ryle, G., The Concept of Mind (Hutchinson, 1949).
Schopenhauer, A., On the Fourfold Root of the Principle of Sufficient Rea-
Unger, P. (1)'An Analysis of Factual Knowledge', \n Journal of Philosophy,
(1968), 157-70.
(2) 'A Defence of Skepticism', in Philosophical Review, (1971),
(3) Ignorancea Case for Scepticism (Oxford, Clarendon Press, 1975).
(4) Philosophical Relativity (Oxford, Basil Blackwell, 1984).
Welbourne, M., The Community of Knowledge (Aberdeen University
Press, 1986).
Williams, B. A. O. (1) 'Knowledge and Reasons', in Problems in the Theory
of Knowledge, ed. von Wright (The Hague, Martinus Nijhoff, 1972),
(2) Descartes: the Project of Pure Enquiry (Penguin, 1978).
This page intentionally left blank

Armstrong, D. M. 29, 102 n. Lehrer, K. 49 n., 77-8

Austin, J. L. 99 Lewis, D. K. 19, 23
Locke,]. 9
Belloc, H. 17
Berkeley, G. 115 McGinn, C. 45,54-9
Blackburn, S. W. 52-33 Moore, G. E. 105, 123

Carnap, R. 8 Nagel, T. 120, 125-9

Craig, E. J. 8n., 105n. Nozick, R. 9, 18-23, 25, 27-9,
34, 45, 54, 56, 58-9, 62, 72, 105
Descartes, R. 86, 104, 108, 116,
118, 120, 124 Pascal, B. 86-7
Dretske, F. 19 n., 27-9, 34, 45, Pears, D. F. 8
62, 72, 163 Plato 7

Edwards, P. 107 n. Quine, W. v. O. 45

Forbes, G. 22 Radford, C. 12, 15-16, 31,

37-8, 40, 82, 137
Gettier, E. 45, 48-9, 53, 69, 71, Ramsey, F. P. 29, 77 n.
75, 77-8,80 Ross, A. 36 n.
God 126 Ryle, G. 153-4
Goldman, A. 24-5, 28-9, 34,
41, 62, 72 Schopenhauer, A. 67
Grice, H. P. 33-4 Socrates 7

Harman, G. 49n., 77 Unger, P. 61, 109-13, 162-7

Hintikka,]. 67
Hobbes, T. 9,84 Welbourne, M. 137,139
Hume, D. 7,9,28,104 Williams, B. A. O. 18,47,72,
75-6, 120, 124-5, 128-9
Kripke, S. 121 Wittgenstein, L. 14, 121